1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public
17 * License along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA.
23 #include <linux/slab.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <asm/byteorder.h>
27 #include <linux/swap.h>
28 #include <linux/pipe_fs_i.h>
29 #include <linux/mpage.h>
31 #define MLOG_MASK_PREFIX ML_FILE_IO
32 #include <cluster/masklog.h>
39 #include "extent_map.h"
47 #include "buffer_head_io.h"
49 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
50 struct buffer_head *bh_result, int create)
54 struct ocfs2_dinode *fe = NULL;
55 struct buffer_head *bh = NULL;
56 struct buffer_head *buffer_cache_bh = NULL;
57 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
60 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
61 (unsigned long long)iblock, bh_result, create);
63 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
65 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
66 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
67 (unsigned long long)iblock);
71 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
72 OCFS2_I(inode)->ip_blkno,
73 &bh, OCFS2_BH_CACHED, inode);
78 fe = (struct ocfs2_dinode *) bh->b_data;
80 if (!OCFS2_IS_VALID_DINODE(fe)) {
81 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
82 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
87 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
88 le32_to_cpu(fe->i_clusters))) {
89 mlog(ML_ERROR, "block offset is outside the allocated size: "
90 "%llu\n", (unsigned long long)iblock);
94 /* We don't use the page cache to create symlink data, so if
95 * need be, copy it over from the buffer cache. */
96 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
97 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
99 buffer_cache_bh = sb_getblk(osb->sb, blkno);
100 if (!buffer_cache_bh) {
101 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
105 /* we haven't locked out transactions, so a commit
106 * could've happened. Since we've got a reference on
107 * the bh, even if it commits while we're doing the
108 * copy, the data is still good. */
109 if (buffer_jbd(buffer_cache_bh)
110 && ocfs2_inode_is_new(inode)) {
111 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
113 mlog(ML_ERROR, "couldn't kmap!\n");
116 memcpy(kaddr + (bh_result->b_size * iblock),
117 buffer_cache_bh->b_data,
119 kunmap_atomic(kaddr, KM_USER0);
120 set_buffer_uptodate(bh_result);
122 brelse(buffer_cache_bh);
125 map_bh(bh_result, inode->i_sb,
126 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
138 static int ocfs2_get_block(struct inode *inode, sector_t iblock,
139 struct buffer_head *bh_result, int create)
142 unsigned int ext_flags;
143 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
144 u64 p_blkno, count, past_eof;
145 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
147 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
148 (unsigned long long)iblock, bh_result, create);
150 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
151 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
152 inode, inode->i_ino);
154 if (S_ISLNK(inode->i_mode)) {
155 /* this always does I/O for some reason. */
156 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
160 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
163 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
164 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
165 (unsigned long long)p_blkno);
169 if (max_blocks < count)
173 * ocfs2 never allocates in this function - the only time we
174 * need to use BH_New is when we're extending i_size on a file
175 * system which doesn't support holes, in which case BH_New
176 * allows block_prepare_write() to zero.
178 mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb),
179 "ino %lu, iblock %llu\n", inode->i_ino,
180 (unsigned long long)iblock);
182 /* Treat the unwritten extent as a hole for zeroing purposes. */
183 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
184 map_bh(bh_result, inode->i_sb, p_blkno);
186 bh_result->b_size = count << inode->i_blkbits;
188 if (!ocfs2_sparse_alloc(osb)) {
192 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
193 (unsigned long long)iblock,
194 (unsigned long long)p_blkno,
195 (unsigned long long)OCFS2_I(inode)->ip_blkno);
196 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
200 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
201 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
202 (unsigned long long)past_eof);
204 if (create && (iblock >= past_eof))
205 set_buffer_new(bh_result);
216 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
217 struct buffer_head *di_bh)
221 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
223 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
224 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
225 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 size = i_size_read(inode);
231 if (size > PAGE_CACHE_SIZE ||
232 size > ocfs2_max_inline_data(inode->i_sb)) {
233 ocfs2_error(inode->i_sb,
234 "Inode %llu has with inline data has bad size: %u",
235 (unsigned long long)OCFS2_I(inode)->ip_blkno, size);
239 kaddr = kmap_atomic(page, KM_USER0);
241 memcpy(kaddr, di->id2.i_data.id_data, size);
242 /* Clear the remaining part of the page */
243 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
244 flush_dcache_page(page);
245 kunmap_atomic(kaddr, KM_USER0);
247 SetPageUptodate(page);
252 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
255 struct buffer_head *di_bh = NULL;
256 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
258 BUG_ON(!PageLocked(page));
259 BUG_ON(!OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL);
261 ret = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &di_bh,
262 OCFS2_BH_CACHED, inode);
268 ret = ocfs2_read_inline_data(inode, page, di_bh);
276 static int ocfs2_readpage(struct file *file, struct page *page)
278 struct inode *inode = page->mapping->host;
279 struct ocfs2_inode_info *oi = OCFS2_I(inode);
280 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
283 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
285 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
287 if (ret == AOP_TRUNCATED_PAGE)
293 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
294 ret = AOP_TRUNCATED_PAGE;
295 goto out_inode_unlock;
299 * i_size might have just been updated as we grabed the meta lock. We
300 * might now be discovering a truncate that hit on another node.
301 * block_read_full_page->get_block freaks out if it is asked to read
302 * beyond the end of a file, so we check here. Callers
303 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
304 * and notice that the page they just read isn't needed.
306 * XXX sys_readahead() seems to get that wrong?
308 if (start >= i_size_read(inode)) {
309 zero_user_page(page, 0, PAGE_SIZE, KM_USER0);
310 SetPageUptodate(page);
315 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
316 ret = ocfs2_readpage_inline(inode, page);
318 ret = block_read_full_page(page, ocfs2_get_block);
322 up_read(&OCFS2_I(inode)->ip_alloc_sem);
324 ocfs2_inode_unlock(inode, 0);
333 * This is used only for read-ahead. Failures or difficult to handle
334 * situations are safe to ignore.
336 * Right now, we don't bother with BH_Boundary - in-inode extent lists
337 * are quite large (243 extents on 4k blocks), so most inodes don't
338 * grow out to a tree. If need be, detecting boundary extents could
339 * trivially be added in a future version of ocfs2_get_block().
341 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
342 struct list_head *pages, unsigned nr_pages)
345 struct inode *inode = mapping->host;
346 struct ocfs2_inode_info *oi = OCFS2_I(inode);
351 * Use the nonblocking flag for the dlm code to avoid page
352 * lock inversion, but don't bother with retrying.
354 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
358 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
359 ocfs2_inode_unlock(inode, 0);
364 * Don't bother with inline-data. There isn't anything
365 * to read-ahead in that case anyway...
367 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
371 * Check whether a remote node truncated this file - we just
372 * drop out in that case as it's not worth handling here.
374 last = list_entry(pages->prev, struct page, lru);
375 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
376 if (start >= i_size_read(inode))
379 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
382 up_read(&oi->ip_alloc_sem);
383 ocfs2_inode_unlock(inode, 0);
388 /* Note: Because we don't support holes, our allocation has
389 * already happened (allocation writes zeros to the file data)
390 * so we don't have to worry about ordered writes in
393 * ->writepage is called during the process of invalidating the page cache
394 * during blocked lock processing. It can't block on any cluster locks
395 * to during block mapping. It's relying on the fact that the block
396 * mapping can't have disappeared under the dirty pages that it is
397 * being asked to write back.
399 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
403 mlog_entry("(0x%p)\n", page);
405 ret = block_write_full_page(page, ocfs2_get_block, wbc);
413 * This is called from ocfs2_write_zero_page() which has handled it's
414 * own cluster locking and has ensured allocation exists for those
415 * blocks to be written.
417 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
418 unsigned from, unsigned to)
422 ret = block_prepare_write(page, from, to, ocfs2_get_block);
427 /* Taken from ext3. We don't necessarily need the full blown
428 * functionality yet, but IMHO it's better to cut and paste the whole
429 * thing so we can avoid introducing our own bugs (and easily pick up
430 * their fixes when they happen) --Mark */
431 int walk_page_buffers( handle_t *handle,
432 struct buffer_head *head,
436 int (*fn)( handle_t *handle,
437 struct buffer_head *bh))
439 struct buffer_head *bh;
440 unsigned block_start, block_end;
441 unsigned blocksize = head->b_size;
443 struct buffer_head *next;
445 for ( bh = head, block_start = 0;
446 ret == 0 && (bh != head || !block_start);
447 block_start = block_end, bh = next)
449 next = bh->b_this_page;
450 block_end = block_start + blocksize;
451 if (block_end <= from || block_start >= to) {
452 if (partial && !buffer_uptodate(bh))
456 err = (*fn)(handle, bh);
463 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
468 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
469 handle_t *handle = NULL;
472 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
479 if (ocfs2_should_order_data(inode)) {
480 ret = walk_page_buffers(handle,
483 ocfs2_journal_dirty_data);
490 ocfs2_commit_trans(osb, handle);
491 handle = ERR_PTR(ret);
496 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
501 struct inode *inode = mapping->host;
503 mlog_entry("(block = %llu)\n", (unsigned long long)block);
505 /* We don't need to lock journal system files, since they aren't
506 * accessed concurrently from multiple nodes.
508 if (!INODE_JOURNAL(inode)) {
509 err = ocfs2_inode_lock(inode, NULL, 0);
515 down_read(&OCFS2_I(inode)->ip_alloc_sem);
518 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
519 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
522 if (!INODE_JOURNAL(inode)) {
523 up_read(&OCFS2_I(inode)->ip_alloc_sem);
524 ocfs2_inode_unlock(inode, 0);
528 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
529 (unsigned long long)block);
535 status = err ? 0 : p_blkno;
537 mlog_exit((int)status);
543 * TODO: Make this into a generic get_blocks function.
545 * From do_direct_io in direct-io.c:
546 * "So what we do is to permit the ->get_blocks function to populate
547 * bh.b_size with the size of IO which is permitted at this offset and
550 * This function is called directly from get_more_blocks in direct-io.c.
552 * called like this: dio->get_blocks(dio->inode, fs_startblk,
553 * fs_count, map_bh, dio->rw == WRITE);
555 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
556 struct buffer_head *bh_result, int create)
559 u64 p_blkno, inode_blocks, contig_blocks;
560 unsigned int ext_flags;
561 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
562 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
564 /* This function won't even be called if the request isn't all
565 * nicely aligned and of the right size, so there's no need
566 * for us to check any of that. */
568 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
571 * Any write past EOF is not allowed because we'd be extending.
573 if (create && (iblock + max_blocks) > inode_blocks) {
578 /* This figures out the size of the next contiguous block, and
579 * our logical offset */
580 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
581 &contig_blocks, &ext_flags);
583 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
584 (unsigned long long)iblock);
589 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) {
590 ocfs2_error(inode->i_sb,
591 "Inode %llu has a hole at block %llu\n",
592 (unsigned long long)OCFS2_I(inode)->ip_blkno,
593 (unsigned long long)iblock);
599 * get_more_blocks() expects us to describe a hole by clearing
600 * the mapped bit on bh_result().
602 * Consider an unwritten extent as a hole.
604 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
605 map_bh(bh_result, inode->i_sb, p_blkno);
608 * ocfs2_prepare_inode_for_write() should have caught
609 * the case where we'd be filling a hole and triggered
610 * a buffered write instead.
618 clear_buffer_mapped(bh_result);
621 /* make sure we don't map more than max_blocks blocks here as
622 that's all the kernel will handle at this point. */
623 if (max_blocks < contig_blocks)
624 contig_blocks = max_blocks;
625 bh_result->b_size = contig_blocks << blocksize_bits;
631 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
632 * particularly interested in the aio/dio case. Like the core uses
633 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
634 * truncation on another.
636 static void ocfs2_dio_end_io(struct kiocb *iocb,
641 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
644 /* this io's submitter should not have unlocked this before we could */
645 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
647 ocfs2_iocb_clear_rw_locked(iocb);
649 level = ocfs2_iocb_rw_locked_level(iocb);
651 up_read(&inode->i_alloc_sem);
652 ocfs2_rw_unlock(inode, level);
656 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
657 * from ext3. PageChecked() bits have been removed as OCFS2 does not
658 * do journalled data.
660 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
662 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
664 journal_invalidatepage(journal, page, offset);
667 static int ocfs2_releasepage(struct page *page, gfp_t wait)
669 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
671 if (!page_has_buffers(page))
673 return journal_try_to_free_buffers(journal, page, wait);
676 static ssize_t ocfs2_direct_IO(int rw,
678 const struct iovec *iov,
680 unsigned long nr_segs)
682 struct file *file = iocb->ki_filp;
683 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
689 * Fallback to buffered I/O if we see an inode without
692 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
695 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
696 inode->i_sb->s_bdev, iov, offset,
698 ocfs2_direct_IO_get_blocks,
705 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
710 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
712 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
715 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
717 cluster_start = cpos % cpp;
718 cluster_start = cluster_start << osb->s_clustersize_bits;
720 cluster_end = cluster_start + osb->s_clustersize;
723 BUG_ON(cluster_start > PAGE_SIZE);
724 BUG_ON(cluster_end > PAGE_SIZE);
727 *start = cluster_start;
733 * 'from' and 'to' are the region in the page to avoid zeroing.
735 * If pagesize > clustersize, this function will avoid zeroing outside
736 * of the cluster boundary.
738 * from == to == 0 is code for "zero the entire cluster region"
740 static void ocfs2_clear_page_regions(struct page *page,
741 struct ocfs2_super *osb, u32 cpos,
742 unsigned from, unsigned to)
745 unsigned int cluster_start, cluster_end;
747 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
749 kaddr = kmap_atomic(page, KM_USER0);
752 if (from > cluster_start)
753 memset(kaddr + cluster_start, 0, from - cluster_start);
754 if (to < cluster_end)
755 memset(kaddr + to, 0, cluster_end - to);
757 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
760 kunmap_atomic(kaddr, KM_USER0);
764 * Nonsparse file systems fully allocate before we get to the write
765 * code. This prevents ocfs2_write() from tagging the write as an
766 * allocating one, which means ocfs2_map_page_blocks() might try to
767 * read-in the blocks at the tail of our file. Avoid reading them by
768 * testing i_size against each block offset.
770 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
771 unsigned int block_start)
773 u64 offset = page_offset(page) + block_start;
775 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
778 if (i_size_read(inode) > offset)
785 * Some of this taken from block_prepare_write(). We already have our
786 * mapping by now though, and the entire write will be allocating or
787 * it won't, so not much need to use BH_New.
789 * This will also skip zeroing, which is handled externally.
791 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
792 struct inode *inode, unsigned int from,
793 unsigned int to, int new)
796 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
797 unsigned int block_end, block_start;
798 unsigned int bsize = 1 << inode->i_blkbits;
800 if (!page_has_buffers(page))
801 create_empty_buffers(page, bsize, 0);
803 head = page_buffers(page);
804 for (bh = head, block_start = 0; bh != head || !block_start;
805 bh = bh->b_this_page, block_start += bsize) {
806 block_end = block_start + bsize;
808 clear_buffer_new(bh);
811 * Ignore blocks outside of our i/o range -
812 * they may belong to unallocated clusters.
814 if (block_start >= to || block_end <= from) {
815 if (PageUptodate(page))
816 set_buffer_uptodate(bh);
821 * For an allocating write with cluster size >= page
822 * size, we always write the entire page.
827 if (!buffer_mapped(bh)) {
828 map_bh(bh, inode->i_sb, *p_blkno);
829 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
832 if (PageUptodate(page)) {
833 if (!buffer_uptodate(bh))
834 set_buffer_uptodate(bh);
835 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
837 ocfs2_should_read_blk(inode, page, block_start) &&
838 (block_start < from || block_end > to)) {
839 ll_rw_block(READ, 1, &bh);
843 *p_blkno = *p_blkno + 1;
847 * If we issued read requests - let them complete.
849 while(wait_bh > wait) {
850 wait_on_buffer(*--wait_bh);
851 if (!buffer_uptodate(*wait_bh))
855 if (ret == 0 || !new)
859 * If we get -EIO above, zero out any newly allocated blocks
860 * to avoid exposing stale data.
865 block_end = block_start + bsize;
866 if (block_end <= from)
868 if (block_start >= to)
871 zero_user_page(page, block_start, bh->b_size, KM_USER0);
872 set_buffer_uptodate(bh);
873 mark_buffer_dirty(bh);
876 block_start = block_end;
877 bh = bh->b_this_page;
878 } while (bh != head);
883 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
884 #define OCFS2_MAX_CTXT_PAGES 1
886 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
889 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
892 * Describe the state of a single cluster to be written to.
894 struct ocfs2_write_cluster_desc {
898 * Give this a unique field because c_phys eventually gets
902 unsigned c_unwritten;
905 static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
907 return d->c_new || d->c_unwritten;
910 struct ocfs2_write_ctxt {
911 /* Logical cluster position / len of write */
915 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
918 * This is true if page_size > cluster_size.
920 * It triggers a set of special cases during write which might
921 * have to deal with allocating writes to partial pages.
923 unsigned int w_large_pages;
926 * Pages involved in this write.
928 * w_target_page is the page being written to by the user.
930 * w_pages is an array of pages which always contains
931 * w_target_page, and in the case of an allocating write with
932 * page_size < cluster size, it will contain zero'd and mapped
933 * pages adjacent to w_target_page which need to be written
934 * out in so that future reads from that region will get
937 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
938 unsigned int w_num_pages;
939 struct page *w_target_page;
942 * ocfs2_write_end() uses this to know what the real range to
943 * write in the target should be.
945 unsigned int w_target_from;
946 unsigned int w_target_to;
949 * We could use journal_current_handle() but this is cleaner,
954 struct buffer_head *w_di_bh;
956 struct ocfs2_cached_dealloc_ctxt w_dealloc;
959 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
963 for(i = 0; i < num_pages; i++) {
965 unlock_page(pages[i]);
966 mark_page_accessed(pages[i]);
967 page_cache_release(pages[i]);
972 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
974 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
980 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
981 struct ocfs2_super *osb, loff_t pos,
982 unsigned len, struct buffer_head *di_bh)
985 struct ocfs2_write_ctxt *wc;
987 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
991 wc->w_cpos = pos >> osb->s_clustersize_bits;
992 cend = (pos + len - 1) >> osb->s_clustersize_bits;
993 wc->w_clen = cend - wc->w_cpos + 1;
997 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
998 wc->w_large_pages = 1;
1000 wc->w_large_pages = 0;
1002 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1010 * If a page has any new buffers, zero them out here, and mark them uptodate
1011 * and dirty so they'll be written out (in order to prevent uninitialised
1012 * block data from leaking). And clear the new bit.
1014 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1016 unsigned int block_start, block_end;
1017 struct buffer_head *head, *bh;
1019 BUG_ON(!PageLocked(page));
1020 if (!page_has_buffers(page))
1023 bh = head = page_buffers(page);
1026 block_end = block_start + bh->b_size;
1028 if (buffer_new(bh)) {
1029 if (block_end > from && block_start < to) {
1030 if (!PageUptodate(page)) {
1031 unsigned start, end;
1033 start = max(from, block_start);
1034 end = min(to, block_end);
1036 zero_user_page(page, start, end - start, KM_USER0);
1037 set_buffer_uptodate(bh);
1040 clear_buffer_new(bh);
1041 mark_buffer_dirty(bh);
1045 block_start = block_end;
1046 bh = bh->b_this_page;
1047 } while (bh != head);
1051 * Only called when we have a failure during allocating write to write
1052 * zero's to the newly allocated region.
1054 static void ocfs2_write_failure(struct inode *inode,
1055 struct ocfs2_write_ctxt *wc,
1056 loff_t user_pos, unsigned user_len)
1059 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1060 to = user_pos + user_len;
1061 struct page *tmppage;
1063 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1065 for(i = 0; i < wc->w_num_pages; i++) {
1066 tmppage = wc->w_pages[i];
1068 if (ocfs2_should_order_data(inode))
1069 walk_page_buffers(wc->w_handle, page_buffers(tmppage),
1071 ocfs2_journal_dirty_data);
1073 block_commit_write(tmppage, from, to);
1077 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1078 struct ocfs2_write_ctxt *wc,
1079 struct page *page, u32 cpos,
1080 loff_t user_pos, unsigned user_len,
1084 unsigned int map_from = 0, map_to = 0;
1085 unsigned int cluster_start, cluster_end;
1086 unsigned int user_data_from = 0, user_data_to = 0;
1088 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1089 &cluster_start, &cluster_end);
1091 if (page == wc->w_target_page) {
1092 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1093 map_to = map_from + user_len;
1096 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1097 cluster_start, cluster_end,
1100 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1101 map_from, map_to, new);
1107 user_data_from = map_from;
1108 user_data_to = map_to;
1110 map_from = cluster_start;
1111 map_to = cluster_end;
1115 * If we haven't allocated the new page yet, we
1116 * shouldn't be writing it out without copying user
1117 * data. This is likely a math error from the caller.
1121 map_from = cluster_start;
1122 map_to = cluster_end;
1124 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1125 cluster_start, cluster_end, new);
1133 * Parts of newly allocated pages need to be zero'd.
1135 * Above, we have also rewritten 'to' and 'from' - as far as
1136 * the rest of the function is concerned, the entire cluster
1137 * range inside of a page needs to be written.
1139 * We can skip this if the page is up to date - it's already
1140 * been zero'd from being read in as a hole.
1142 if (new && !PageUptodate(page))
1143 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1144 cpos, user_data_from, user_data_to);
1146 flush_dcache_page(page);
1153 * This function will only grab one clusters worth of pages.
1155 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1156 struct ocfs2_write_ctxt *wc,
1157 u32 cpos, loff_t user_pos, int new,
1158 struct page *mmap_page)
1161 unsigned long start, target_index, index;
1162 struct inode *inode = mapping->host;
1164 target_index = user_pos >> PAGE_CACHE_SHIFT;
1167 * Figure out how many pages we'll be manipulating here. For
1168 * non allocating write, we just change the one
1169 * page. Otherwise, we'll need a whole clusters worth.
1172 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1173 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1175 wc->w_num_pages = 1;
1176 start = target_index;
1179 for(i = 0; i < wc->w_num_pages; i++) {
1182 if (index == target_index && mmap_page) {
1184 * ocfs2_pagemkwrite() is a little different
1185 * and wants us to directly use the page
1188 lock_page(mmap_page);
1190 if (mmap_page->mapping != mapping) {
1191 unlock_page(mmap_page);
1193 * Sanity check - the locking in
1194 * ocfs2_pagemkwrite() should ensure
1195 * that this code doesn't trigger.
1202 page_cache_get(mmap_page);
1203 wc->w_pages[i] = mmap_page;
1205 wc->w_pages[i] = find_or_create_page(mapping, index,
1207 if (!wc->w_pages[i]) {
1214 if (index == target_index)
1215 wc->w_target_page = wc->w_pages[i];
1222 * Prepare a single cluster for write one cluster into the file.
1224 static int ocfs2_write_cluster(struct address_space *mapping,
1225 u32 phys, unsigned int unwritten,
1226 struct ocfs2_alloc_context *data_ac,
1227 struct ocfs2_alloc_context *meta_ac,
1228 struct ocfs2_write_ctxt *wc, u32 cpos,
1229 loff_t user_pos, unsigned user_len)
1231 int ret, i, new, should_zero = 0;
1232 u64 v_blkno, p_blkno;
1233 struct inode *inode = mapping->host;
1235 new = phys == 0 ? 1 : 0;
1236 if (new || unwritten)
1243 * This is safe to call with the page locks - it won't take
1244 * any additional semaphores or cluster locks.
1247 ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
1248 &tmp_pos, 1, 0, wc->w_di_bh,
1249 wc->w_handle, data_ac,
1252 * This shouldn't happen because we must have already
1253 * calculated the correct meta data allocation required. The
1254 * internal tree allocation code should know how to increase
1255 * transaction credits itself.
1257 * If need be, we could handle -EAGAIN for a
1258 * RESTART_TRANS here.
1260 mlog_bug_on_msg(ret == -EAGAIN,
1261 "Inode %llu: EAGAIN return during allocation.\n",
1262 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1267 } else if (unwritten) {
1268 ret = ocfs2_mark_extent_written(inode, wc->w_di_bh,
1269 wc->w_handle, cpos, 1, phys,
1270 meta_ac, &wc->w_dealloc);
1278 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1280 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1283 * The only reason this should fail is due to an inability to
1284 * find the extent added.
1286 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1289 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1290 "at logical block %llu",
1291 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1292 (unsigned long long)v_blkno);
1296 BUG_ON(p_blkno == 0);
1298 for(i = 0; i < wc->w_num_pages; i++) {
1301 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1302 wc->w_pages[i], cpos,
1313 * We only have cleanup to do in case of allocating write.
1316 ocfs2_write_failure(inode, wc, user_pos, user_len);
1323 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1324 struct ocfs2_alloc_context *data_ac,
1325 struct ocfs2_alloc_context *meta_ac,
1326 struct ocfs2_write_ctxt *wc,
1327 loff_t pos, unsigned len)
1331 unsigned int local_len = len;
1332 struct ocfs2_write_cluster_desc *desc;
1333 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1335 for (i = 0; i < wc->w_clen; i++) {
1336 desc = &wc->w_desc[i];
1339 * We have to make sure that the total write passed in
1340 * doesn't extend past a single cluster.
1343 cluster_off = pos & (osb->s_clustersize - 1);
1344 if ((cluster_off + local_len) > osb->s_clustersize)
1345 local_len = osb->s_clustersize - cluster_off;
1347 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1348 desc->c_unwritten, data_ac, meta_ac,
1349 wc, desc->c_cpos, pos, local_len);
1365 * ocfs2_write_end() wants to know which parts of the target page it
1366 * should complete the write on. It's easiest to compute them ahead of
1367 * time when a more complete view of the write is available.
1369 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1370 struct ocfs2_write_ctxt *wc,
1371 loff_t pos, unsigned len, int alloc)
1373 struct ocfs2_write_cluster_desc *desc;
1375 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1376 wc->w_target_to = wc->w_target_from + len;
1382 * Allocating write - we may have different boundaries based
1383 * on page size and cluster size.
1385 * NOTE: We can no longer compute one value from the other as
1386 * the actual write length and user provided length may be
1390 if (wc->w_large_pages) {
1392 * We only care about the 1st and last cluster within
1393 * our range and whether they should be zero'd or not. Either
1394 * value may be extended out to the start/end of a
1395 * newly allocated cluster.
1397 desc = &wc->w_desc[0];
1398 if (ocfs2_should_zero_cluster(desc))
1399 ocfs2_figure_cluster_boundaries(osb,
1404 desc = &wc->w_desc[wc->w_clen - 1];
1405 if (ocfs2_should_zero_cluster(desc))
1406 ocfs2_figure_cluster_boundaries(osb,
1411 wc->w_target_from = 0;
1412 wc->w_target_to = PAGE_CACHE_SIZE;
1417 * Populate each single-cluster write descriptor in the write context
1418 * with information about the i/o to be done.
1420 * Returns the number of clusters that will have to be allocated, as
1421 * well as a worst case estimate of the number of extent records that
1422 * would have to be created during a write to an unwritten region.
1424 static int ocfs2_populate_write_desc(struct inode *inode,
1425 struct ocfs2_write_ctxt *wc,
1426 unsigned int *clusters_to_alloc,
1427 unsigned int *extents_to_split)
1430 struct ocfs2_write_cluster_desc *desc;
1431 unsigned int num_clusters = 0;
1432 unsigned int ext_flags = 0;
1436 *clusters_to_alloc = 0;
1437 *extents_to_split = 0;
1439 for (i = 0; i < wc->w_clen; i++) {
1440 desc = &wc->w_desc[i];
1441 desc->c_cpos = wc->w_cpos + i;
1443 if (num_clusters == 0) {
1445 * Need to look up the next extent record.
1447 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1448 &num_clusters, &ext_flags);
1455 * Assume worst case - that we're writing in
1456 * the middle of the extent.
1458 * We can assume that the write proceeds from
1459 * left to right, in which case the extent
1460 * insert code is smart enough to coalesce the
1461 * next splits into the previous records created.
1463 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1464 *extents_to_split = *extents_to_split + 2;
1467 * Only increment phys if it doesn't describe
1473 desc->c_phys = phys;
1476 *clusters_to_alloc = *clusters_to_alloc + 1;
1478 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1479 desc->c_unwritten = 1;
1489 static int ocfs2_write_begin_inline(struct address_space *mapping,
1490 struct inode *inode,
1491 struct ocfs2_write_ctxt *wc)
1494 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1497 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1499 page = find_or_create_page(mapping, 0, GFP_NOFS);
1506 * If we don't set w_num_pages then this page won't get unlocked
1507 * and freed on cleanup of the write context.
1509 wc->w_pages[0] = wc->w_target_page = page;
1510 wc->w_num_pages = 1;
1512 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1513 if (IS_ERR(handle)) {
1514 ret = PTR_ERR(handle);
1519 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1520 OCFS2_JOURNAL_ACCESS_WRITE);
1522 ocfs2_commit_trans(osb, handle);
1528 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1529 ocfs2_set_inode_data_inline(inode, di);
1531 if (!PageUptodate(page)) {
1532 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1534 ocfs2_commit_trans(osb, handle);
1540 wc->w_handle = handle;
1545 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1547 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1549 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1554 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1555 struct inode *inode, loff_t pos,
1556 unsigned len, struct page *mmap_page,
1557 struct ocfs2_write_ctxt *wc)
1559 int ret, written = 0;
1560 loff_t end = pos + len;
1561 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1563 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1564 (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos,
1565 oi->ip_dyn_features);
1568 * Handle inodes which already have inline data 1st.
1570 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1571 if (mmap_page == NULL &&
1572 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1573 goto do_inline_write;
1576 * The write won't fit - we have to give this inode an
1577 * inline extent list now.
1579 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1586 * Check whether the inode can accept inline data.
1588 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1592 * Check whether the write can fit.
1594 if (mmap_page || end > ocfs2_max_inline_data(inode->i_sb))
1598 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1605 * This signals to the caller that the data can be written
1610 return written ? written : ret;
1614 * This function only does anything for file systems which can't
1615 * handle sparse files.
1617 * What we want to do here is fill in any hole between the current end
1618 * of allocation and the end of our write. That way the rest of the
1619 * write path can treat it as an non-allocating write, which has no
1620 * special case code for sparse/nonsparse files.
1622 static int ocfs2_expand_nonsparse_inode(struct inode *inode, loff_t pos,
1624 struct ocfs2_write_ctxt *wc)
1627 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1628 loff_t newsize = pos + len;
1630 if (ocfs2_sparse_alloc(osb))
1633 if (newsize <= i_size_read(inode))
1636 ret = ocfs2_extend_no_holes(inode, newsize, newsize - len);
1643 int ocfs2_write_begin_nolock(struct address_space *mapping,
1644 loff_t pos, unsigned len, unsigned flags,
1645 struct page **pagep, void **fsdata,
1646 struct buffer_head *di_bh, struct page *mmap_page)
1648 int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
1649 unsigned int clusters_to_alloc, extents_to_split;
1650 struct ocfs2_write_ctxt *wc;
1651 struct inode *inode = mapping->host;
1652 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1653 struct ocfs2_dinode *di;
1654 struct ocfs2_alloc_context *data_ac = NULL;
1655 struct ocfs2_alloc_context *meta_ac = NULL;
1658 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1664 if (ocfs2_supports_inline_data(osb)) {
1665 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1677 ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc);
1683 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1690 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1693 * We set w_target_from, w_target_to here so that
1694 * ocfs2_write_end() knows which range in the target page to
1695 * write out. An allocation requires that we write the entire
1698 if (clusters_to_alloc || extents_to_split) {
1700 * XXX: We are stretching the limits of
1701 * ocfs2_lock_allocators(). It greatly over-estimates
1702 * the work to be done.
1704 ret = ocfs2_lock_allocators(inode, di, clusters_to_alloc,
1705 extents_to_split, &data_ac, &meta_ac);
1711 credits = ocfs2_calc_extend_credits(inode->i_sb, di,
1716 ocfs2_set_target_boundaries(osb, wc, pos, len,
1717 clusters_to_alloc + extents_to_split);
1719 handle = ocfs2_start_trans(osb, credits);
1720 if (IS_ERR(handle)) {
1721 ret = PTR_ERR(handle);
1726 wc->w_handle = handle;
1729 * We don't want this to fail in ocfs2_write_end(), so do it
1732 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1733 OCFS2_JOURNAL_ACCESS_WRITE);
1740 * Fill our page array first. That way we've grabbed enough so
1741 * that we can zero and flush if we error after adding the
1744 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
1745 clusters_to_alloc + extents_to_split,
1752 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1760 ocfs2_free_alloc_context(data_ac);
1762 ocfs2_free_alloc_context(meta_ac);
1765 *pagep = wc->w_target_page;
1769 ocfs2_commit_trans(osb, handle);
1772 ocfs2_free_write_ctxt(wc);
1775 ocfs2_free_alloc_context(data_ac);
1777 ocfs2_free_alloc_context(meta_ac);
1781 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1782 loff_t pos, unsigned len, unsigned flags,
1783 struct page **pagep, void **fsdata)
1786 struct buffer_head *di_bh = NULL;
1787 struct inode *inode = mapping->host;
1789 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1796 * Take alloc sem here to prevent concurrent lookups. That way
1797 * the mapping, zeroing and tree manipulation within
1798 * ocfs2_write() will be safe against ->readpage(). This
1799 * should also serve to lock out allocation from a shared
1802 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1804 ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
1805 fsdata, di_bh, NULL);
1816 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1819 ocfs2_inode_unlock(inode, 1);
1824 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1825 unsigned len, unsigned *copied,
1826 struct ocfs2_dinode *di,
1827 struct ocfs2_write_ctxt *wc)
1831 if (unlikely(*copied < len)) {
1832 if (!PageUptodate(wc->w_target_page)) {
1838 kaddr = kmap_atomic(wc->w_target_page, KM_USER0);
1839 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1840 kunmap_atomic(kaddr, KM_USER0);
1842 mlog(0, "Data written to inode at offset %llu. "
1843 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1844 (unsigned long long)pos, *copied,
1845 le16_to_cpu(di->id2.i_data.id_count),
1846 le16_to_cpu(di->i_dyn_features));
1849 int ocfs2_write_end_nolock(struct address_space *mapping,
1850 loff_t pos, unsigned len, unsigned copied,
1851 struct page *page, void *fsdata)
1854 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1855 struct inode *inode = mapping->host;
1856 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1857 struct ocfs2_write_ctxt *wc = fsdata;
1858 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1859 handle_t *handle = wc->w_handle;
1860 struct page *tmppage;
1862 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1863 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1864 goto out_write_size;
1867 if (unlikely(copied < len)) {
1868 if (!PageUptodate(wc->w_target_page))
1871 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1874 flush_dcache_page(wc->w_target_page);
1876 for(i = 0; i < wc->w_num_pages; i++) {
1877 tmppage = wc->w_pages[i];
1879 if (tmppage == wc->w_target_page) {
1880 from = wc->w_target_from;
1881 to = wc->w_target_to;
1883 BUG_ON(from > PAGE_CACHE_SIZE ||
1884 to > PAGE_CACHE_SIZE ||
1888 * Pages adjacent to the target (if any) imply
1889 * a hole-filling write in which case we want
1890 * to flush their entire range.
1893 to = PAGE_CACHE_SIZE;
1896 if (ocfs2_should_order_data(inode))
1897 walk_page_buffers(wc->w_handle, page_buffers(tmppage),
1899 ocfs2_journal_dirty_data);
1901 block_commit_write(tmppage, from, to);
1906 if (pos > inode->i_size) {
1907 i_size_write(inode, pos);
1908 mark_inode_dirty(inode);
1910 inode->i_blocks = ocfs2_inode_sector_count(inode);
1911 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1912 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1913 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1914 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1915 ocfs2_journal_dirty(handle, wc->w_di_bh);
1917 ocfs2_commit_trans(osb, handle);
1919 ocfs2_run_deallocs(osb, &wc->w_dealloc);
1921 ocfs2_free_write_ctxt(wc);
1926 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
1927 loff_t pos, unsigned len, unsigned copied,
1928 struct page *page, void *fsdata)
1931 struct inode *inode = mapping->host;
1933 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
1935 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1936 ocfs2_inode_unlock(inode, 1);
1941 const struct address_space_operations ocfs2_aops = {
1942 .readpage = ocfs2_readpage,
1943 .readpages = ocfs2_readpages,
1944 .writepage = ocfs2_writepage,
1945 .write_begin = ocfs2_write_begin,
1946 .write_end = ocfs2_write_end,
1948 .sync_page = block_sync_page,
1949 .direct_IO = ocfs2_direct_IO,
1950 .invalidatepage = ocfs2_invalidatepage,
1951 .releasepage = ocfs2_releasepage,
1952 .migratepage = buffer_migrate_page,