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_inode_block(inode, &bh);
76 fe = (struct ocfs2_dinode *) bh->b_data;
78 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
79 le32_to_cpu(fe->i_clusters))) {
80 mlog(ML_ERROR, "block offset is outside the allocated size: "
81 "%llu\n", (unsigned long long)iblock);
85 /* We don't use the page cache to create symlink data, so if
86 * need be, copy it over from the buffer cache. */
87 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
88 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
90 buffer_cache_bh = sb_getblk(osb->sb, blkno);
91 if (!buffer_cache_bh) {
92 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
96 /* we haven't locked out transactions, so a commit
97 * could've happened. Since we've got a reference on
98 * the bh, even if it commits while we're doing the
99 * copy, the data is still good. */
100 if (buffer_jbd(buffer_cache_bh)
101 && ocfs2_inode_is_new(inode)) {
102 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
104 mlog(ML_ERROR, "couldn't kmap!\n");
107 memcpy(kaddr + (bh_result->b_size * iblock),
108 buffer_cache_bh->b_data,
110 kunmap_atomic(kaddr, KM_USER0);
111 set_buffer_uptodate(bh_result);
113 brelse(buffer_cache_bh);
116 map_bh(bh_result, inode->i_sb,
117 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
128 static int ocfs2_get_block(struct inode *inode, sector_t iblock,
129 struct buffer_head *bh_result, int create)
132 unsigned int ext_flags;
133 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
134 u64 p_blkno, count, past_eof;
135 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
137 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
138 (unsigned long long)iblock, bh_result, create);
140 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
141 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
142 inode, inode->i_ino);
144 if (S_ISLNK(inode->i_mode)) {
145 /* this always does I/O for some reason. */
146 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
150 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
153 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
154 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
155 (unsigned long long)p_blkno);
159 if (max_blocks < count)
163 * ocfs2 never allocates in this function - the only time we
164 * need to use BH_New is when we're extending i_size on a file
165 * system which doesn't support holes, in which case BH_New
166 * allows block_prepare_write() to zero.
168 * If we see this on a sparse file system, then a truncate has
169 * raced us and removed the cluster. In this case, we clear
170 * the buffers dirty and uptodate bits and let the buffer code
171 * ignore it as a hole.
173 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
174 clear_buffer_dirty(bh_result);
175 clear_buffer_uptodate(bh_result);
179 /* Treat the unwritten extent as a hole for zeroing purposes. */
180 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
181 map_bh(bh_result, inode->i_sb, p_blkno);
183 bh_result->b_size = count << inode->i_blkbits;
185 if (!ocfs2_sparse_alloc(osb)) {
189 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
190 (unsigned long long)iblock,
191 (unsigned long long)p_blkno,
192 (unsigned long long)OCFS2_I(inode)->ip_blkno);
193 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
197 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
198 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
199 (unsigned long long)past_eof);
201 if (create && (iblock >= past_eof))
202 set_buffer_new(bh_result);
213 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
214 struct buffer_head *di_bh)
218 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
220 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
221 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
222 (unsigned long long)OCFS2_I(inode)->ip_blkno);
226 size = i_size_read(inode);
228 if (size > PAGE_CACHE_SIZE ||
229 size > ocfs2_max_inline_data(inode->i_sb)) {
230 ocfs2_error(inode->i_sb,
231 "Inode %llu has with inline data has bad size: %Lu",
232 (unsigned long long)OCFS2_I(inode)->ip_blkno,
233 (unsigned long long)size);
237 kaddr = kmap_atomic(page, KM_USER0);
239 memcpy(kaddr, di->id2.i_data.id_data, size);
240 /* Clear the remaining part of the page */
241 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
242 flush_dcache_page(page);
243 kunmap_atomic(kaddr, KM_USER0);
245 SetPageUptodate(page);
250 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
253 struct buffer_head *di_bh = NULL;
255 BUG_ON(!PageLocked(page));
256 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
258 ret = ocfs2_read_inode_block(inode, &di_bh);
264 ret = ocfs2_read_inline_data(inode, page, di_bh);
272 static int ocfs2_readpage(struct file *file, struct page *page)
274 struct inode *inode = page->mapping->host;
275 struct ocfs2_inode_info *oi = OCFS2_I(inode);
276 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
279 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
281 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
283 if (ret == AOP_TRUNCATED_PAGE)
289 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
290 ret = AOP_TRUNCATED_PAGE;
291 goto out_inode_unlock;
295 * i_size might have just been updated as we grabed the meta lock. We
296 * might now be discovering a truncate that hit on another node.
297 * block_read_full_page->get_block freaks out if it is asked to read
298 * beyond the end of a file, so we check here. Callers
299 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
300 * and notice that the page they just read isn't needed.
302 * XXX sys_readahead() seems to get that wrong?
304 if (start >= i_size_read(inode)) {
305 zero_user(page, 0, PAGE_SIZE);
306 SetPageUptodate(page);
311 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
312 ret = ocfs2_readpage_inline(inode, page);
314 ret = block_read_full_page(page, ocfs2_get_block);
318 up_read(&OCFS2_I(inode)->ip_alloc_sem);
320 ocfs2_inode_unlock(inode, 0);
329 * This is used only for read-ahead. Failures or difficult to handle
330 * situations are safe to ignore.
332 * Right now, we don't bother with BH_Boundary - in-inode extent lists
333 * are quite large (243 extents on 4k blocks), so most inodes don't
334 * grow out to a tree. If need be, detecting boundary extents could
335 * trivially be added in a future version of ocfs2_get_block().
337 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
338 struct list_head *pages, unsigned nr_pages)
341 struct inode *inode = mapping->host;
342 struct ocfs2_inode_info *oi = OCFS2_I(inode);
347 * Use the nonblocking flag for the dlm code to avoid page
348 * lock inversion, but don't bother with retrying.
350 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
354 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
355 ocfs2_inode_unlock(inode, 0);
360 * Don't bother with inline-data. There isn't anything
361 * to read-ahead in that case anyway...
363 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
367 * Check whether a remote node truncated this file - we just
368 * drop out in that case as it's not worth handling here.
370 last = list_entry(pages->prev, struct page, lru);
371 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
372 if (start >= i_size_read(inode))
375 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
378 up_read(&oi->ip_alloc_sem);
379 ocfs2_inode_unlock(inode, 0);
384 /* Note: Because we don't support holes, our allocation has
385 * already happened (allocation writes zeros to the file data)
386 * so we don't have to worry about ordered writes in
389 * ->writepage is called during the process of invalidating the page cache
390 * during blocked lock processing. It can't block on any cluster locks
391 * to during block mapping. It's relying on the fact that the block
392 * mapping can't have disappeared under the dirty pages that it is
393 * being asked to write back.
395 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
399 mlog_entry("(0x%p)\n", page);
401 ret = block_write_full_page(page, ocfs2_get_block, wbc);
409 * This is called from ocfs2_write_zero_page() which has handled it's
410 * own cluster locking and has ensured allocation exists for those
411 * blocks to be written.
413 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
414 unsigned from, unsigned to)
418 ret = block_prepare_write(page, from, to, ocfs2_get_block);
423 /* Taken from ext3. We don't necessarily need the full blown
424 * functionality yet, but IMHO it's better to cut and paste the whole
425 * thing so we can avoid introducing our own bugs (and easily pick up
426 * their fixes when they happen) --Mark */
427 int walk_page_buffers( handle_t *handle,
428 struct buffer_head *head,
432 int (*fn)( handle_t *handle,
433 struct buffer_head *bh))
435 struct buffer_head *bh;
436 unsigned block_start, block_end;
437 unsigned blocksize = head->b_size;
439 struct buffer_head *next;
441 for ( bh = head, block_start = 0;
442 ret == 0 && (bh != head || !block_start);
443 block_start = block_end, bh = next)
445 next = bh->b_this_page;
446 block_end = block_start + blocksize;
447 if (block_end <= from || block_start >= to) {
448 if (partial && !buffer_uptodate(bh))
452 err = (*fn)(handle, bh);
459 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
464 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
468 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
469 if (IS_ERR(handle)) {
475 if (ocfs2_should_order_data(inode)) {
476 ret = ocfs2_jbd2_file_inode(handle, inode);
477 #ifdef CONFIG_OCFS2_COMPAT_JBD
478 ret = walk_page_buffers(handle,
481 ocfs2_journal_dirty_data);
489 ocfs2_commit_trans(osb, handle);
490 handle = ERR_PTR(ret);
495 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
500 struct inode *inode = mapping->host;
502 mlog_entry("(block = %llu)\n", (unsigned long long)block);
504 /* We don't need to lock journal system files, since they aren't
505 * accessed concurrently from multiple nodes.
507 if (!INODE_JOURNAL(inode)) {
508 err = ocfs2_inode_lock(inode, NULL, 0);
514 down_read(&OCFS2_I(inode)->ip_alloc_sem);
517 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
518 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
521 if (!INODE_JOURNAL(inode)) {
522 up_read(&OCFS2_I(inode)->ip_alloc_sem);
523 ocfs2_inode_unlock(inode, 0);
527 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
528 (unsigned long long)block);
534 status = err ? 0 : p_blkno;
536 mlog_exit((int)status);
542 * TODO: Make this into a generic get_blocks function.
544 * From do_direct_io in direct-io.c:
545 * "So what we do is to permit the ->get_blocks function to populate
546 * bh.b_size with the size of IO which is permitted at this offset and
549 * This function is called directly from get_more_blocks in direct-io.c.
551 * called like this: dio->get_blocks(dio->inode, fs_startblk,
552 * fs_count, map_bh, dio->rw == WRITE);
554 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
555 struct buffer_head *bh_result, int create)
558 u64 p_blkno, inode_blocks, contig_blocks;
559 unsigned int ext_flags;
560 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
561 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
563 /* This function won't even be called if the request isn't all
564 * nicely aligned and of the right size, so there's no need
565 * for us to check any of that. */
567 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
570 * Any write past EOF is not allowed because we'd be extending.
572 if (create && (iblock + max_blocks) > inode_blocks) {
577 /* This figures out the size of the next contiguous block, and
578 * our logical offset */
579 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
580 &contig_blocks, &ext_flags);
582 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
583 (unsigned long long)iblock);
588 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno && create) {
589 ocfs2_error(inode->i_sb,
590 "Inode %llu has a hole at block %llu\n",
591 (unsigned long long)OCFS2_I(inode)->ip_blkno,
592 (unsigned long long)iblock);
598 * get_more_blocks() expects us to describe a hole by clearing
599 * the mapped bit on bh_result().
601 * Consider an unwritten extent as a hole.
603 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
604 map_bh(bh_result, inode->i_sb, p_blkno);
607 * ocfs2_prepare_inode_for_write() should have caught
608 * the case where we'd be filling a hole and triggered
609 * a buffered write instead.
617 clear_buffer_mapped(bh_result);
620 /* make sure we don't map more than max_blocks blocks here as
621 that's all the kernel will handle at this point. */
622 if (max_blocks < contig_blocks)
623 contig_blocks = max_blocks;
624 bh_result->b_size = contig_blocks << blocksize_bits;
630 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
631 * particularly interested in the aio/dio case. Like the core uses
632 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
633 * truncation on another.
635 static void ocfs2_dio_end_io(struct kiocb *iocb,
640 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
643 /* this io's submitter should not have unlocked this before we could */
644 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
646 ocfs2_iocb_clear_rw_locked(iocb);
648 level = ocfs2_iocb_rw_locked_level(iocb);
650 up_read(&inode->i_alloc_sem);
651 ocfs2_rw_unlock(inode, level);
655 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
656 * from ext3. PageChecked() bits have been removed as OCFS2 does not
657 * do journalled data.
659 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
661 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
663 jbd2_journal_invalidatepage(journal, page, offset);
666 static int ocfs2_releasepage(struct page *page, gfp_t wait)
668 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
670 if (!page_has_buffers(page))
672 return jbd2_journal_try_to_free_buffers(journal, page, wait);
675 static ssize_t ocfs2_direct_IO(int rw,
677 const struct iovec *iov,
679 unsigned long nr_segs)
681 struct file *file = iocb->ki_filp;
682 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
688 * Fallback to buffered I/O if we see an inode without
691 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
694 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
695 inode->i_sb->s_bdev, iov, offset,
697 ocfs2_direct_IO_get_blocks,
704 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
709 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
711 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
714 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
716 cluster_start = cpos % cpp;
717 cluster_start = cluster_start << osb->s_clustersize_bits;
719 cluster_end = cluster_start + osb->s_clustersize;
722 BUG_ON(cluster_start > PAGE_SIZE);
723 BUG_ON(cluster_end > PAGE_SIZE);
726 *start = cluster_start;
732 * 'from' and 'to' are the region in the page to avoid zeroing.
734 * If pagesize > clustersize, this function will avoid zeroing outside
735 * of the cluster boundary.
737 * from == to == 0 is code for "zero the entire cluster region"
739 static void ocfs2_clear_page_regions(struct page *page,
740 struct ocfs2_super *osb, u32 cpos,
741 unsigned from, unsigned to)
744 unsigned int cluster_start, cluster_end;
746 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
748 kaddr = kmap_atomic(page, KM_USER0);
751 if (from > cluster_start)
752 memset(kaddr + cluster_start, 0, from - cluster_start);
753 if (to < cluster_end)
754 memset(kaddr + to, 0, cluster_end - to);
756 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
759 kunmap_atomic(kaddr, KM_USER0);
763 * Nonsparse file systems fully allocate before we get to the write
764 * code. This prevents ocfs2_write() from tagging the write as an
765 * allocating one, which means ocfs2_map_page_blocks() might try to
766 * read-in the blocks at the tail of our file. Avoid reading them by
767 * testing i_size against each block offset.
769 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
770 unsigned int block_start)
772 u64 offset = page_offset(page) + block_start;
774 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
777 if (i_size_read(inode) > offset)
784 * Some of this taken from block_prepare_write(). We already have our
785 * mapping by now though, and the entire write will be allocating or
786 * it won't, so not much need to use BH_New.
788 * This will also skip zeroing, which is handled externally.
790 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
791 struct inode *inode, unsigned int from,
792 unsigned int to, int new)
795 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
796 unsigned int block_end, block_start;
797 unsigned int bsize = 1 << inode->i_blkbits;
799 if (!page_has_buffers(page))
800 create_empty_buffers(page, bsize, 0);
802 head = page_buffers(page);
803 for (bh = head, block_start = 0; bh != head || !block_start;
804 bh = bh->b_this_page, block_start += bsize) {
805 block_end = block_start + bsize;
807 clear_buffer_new(bh);
810 * Ignore blocks outside of our i/o range -
811 * they may belong to unallocated clusters.
813 if (block_start >= to || block_end <= from) {
814 if (PageUptodate(page))
815 set_buffer_uptodate(bh);
820 * For an allocating write with cluster size >= page
821 * size, we always write the entire page.
826 if (!buffer_mapped(bh)) {
827 map_bh(bh, inode->i_sb, *p_blkno);
828 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
831 if (PageUptodate(page)) {
832 if (!buffer_uptodate(bh))
833 set_buffer_uptodate(bh);
834 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
836 ocfs2_should_read_blk(inode, page, block_start) &&
837 (block_start < from || block_end > to)) {
838 ll_rw_block(READ, 1, &bh);
842 *p_blkno = *p_blkno + 1;
846 * If we issued read requests - let them complete.
848 while(wait_bh > wait) {
849 wait_on_buffer(*--wait_bh);
850 if (!buffer_uptodate(*wait_bh))
854 if (ret == 0 || !new)
858 * If we get -EIO above, zero out any newly allocated blocks
859 * to avoid exposing stale data.
864 block_end = block_start + bsize;
865 if (block_end <= from)
867 if (block_start >= to)
870 zero_user(page, block_start, bh->b_size);
871 set_buffer_uptodate(bh);
872 mark_buffer_dirty(bh);
875 block_start = block_end;
876 bh = bh->b_this_page;
877 } while (bh != head);
882 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
883 #define OCFS2_MAX_CTXT_PAGES 1
885 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
888 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
891 * Describe the state of a single cluster to be written to.
893 struct ocfs2_write_cluster_desc {
897 * Give this a unique field because c_phys eventually gets
901 unsigned c_unwritten;
904 static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
906 return d->c_new || d->c_unwritten;
909 struct ocfs2_write_ctxt {
910 /* Logical cluster position / len of write */
914 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
917 * This is true if page_size > cluster_size.
919 * It triggers a set of special cases during write which might
920 * have to deal with allocating writes to partial pages.
922 unsigned int w_large_pages;
925 * Pages involved in this write.
927 * w_target_page is the page being written to by the user.
929 * w_pages is an array of pages which always contains
930 * w_target_page, and in the case of an allocating write with
931 * page_size < cluster size, it will contain zero'd and mapped
932 * pages adjacent to w_target_page which need to be written
933 * out in so that future reads from that region will get
936 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
937 unsigned int w_num_pages;
938 struct page *w_target_page;
941 * ocfs2_write_end() uses this to know what the real range to
942 * write in the target should be.
944 unsigned int w_target_from;
945 unsigned int w_target_to;
948 * We could use journal_current_handle() but this is cleaner,
953 struct buffer_head *w_di_bh;
955 struct ocfs2_cached_dealloc_ctxt w_dealloc;
958 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
962 for(i = 0; i < num_pages; i++) {
964 unlock_page(pages[i]);
965 mark_page_accessed(pages[i]);
966 page_cache_release(pages[i]);
971 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
973 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
979 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
980 struct ocfs2_super *osb, loff_t pos,
981 unsigned len, struct buffer_head *di_bh)
984 struct ocfs2_write_ctxt *wc;
986 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
990 wc->w_cpos = pos >> osb->s_clustersize_bits;
991 cend = (pos + len - 1) >> osb->s_clustersize_bits;
992 wc->w_clen = cend - wc->w_cpos + 1;
996 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
997 wc->w_large_pages = 1;
999 wc->w_large_pages = 0;
1001 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1009 * If a page has any new buffers, zero them out here, and mark them uptodate
1010 * and dirty so they'll be written out (in order to prevent uninitialised
1011 * block data from leaking). And clear the new bit.
1013 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1015 unsigned int block_start, block_end;
1016 struct buffer_head *head, *bh;
1018 BUG_ON(!PageLocked(page));
1019 if (!page_has_buffers(page))
1022 bh = head = page_buffers(page);
1025 block_end = block_start + bh->b_size;
1027 if (buffer_new(bh)) {
1028 if (block_end > from && block_start < to) {
1029 if (!PageUptodate(page)) {
1030 unsigned start, end;
1032 start = max(from, block_start);
1033 end = min(to, block_end);
1035 zero_user_segment(page, start, end);
1036 set_buffer_uptodate(bh);
1039 clear_buffer_new(bh);
1040 mark_buffer_dirty(bh);
1044 block_start = block_end;
1045 bh = bh->b_this_page;
1046 } while (bh != head);
1050 * Only called when we have a failure during allocating write to write
1051 * zero's to the newly allocated region.
1053 static void ocfs2_write_failure(struct inode *inode,
1054 struct ocfs2_write_ctxt *wc,
1055 loff_t user_pos, unsigned user_len)
1058 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1059 to = user_pos + user_len;
1060 struct page *tmppage;
1062 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1064 for(i = 0; i < wc->w_num_pages; i++) {
1065 tmppage = wc->w_pages[i];
1067 if (page_has_buffers(tmppage)) {
1068 if (ocfs2_should_order_data(inode)) {
1069 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1070 #ifdef CONFIG_OCFS2_COMPAT_JBD
1071 walk_page_buffers(wc->w_handle,
1072 page_buffers(tmppage),
1074 ocfs2_journal_dirty_data);
1078 block_commit_write(tmppage, from, to);
1083 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1084 struct ocfs2_write_ctxt *wc,
1085 struct page *page, u32 cpos,
1086 loff_t user_pos, unsigned user_len,
1090 unsigned int map_from = 0, map_to = 0;
1091 unsigned int cluster_start, cluster_end;
1092 unsigned int user_data_from = 0, user_data_to = 0;
1094 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1095 &cluster_start, &cluster_end);
1097 if (page == wc->w_target_page) {
1098 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1099 map_to = map_from + user_len;
1102 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1103 cluster_start, cluster_end,
1106 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1107 map_from, map_to, new);
1113 user_data_from = map_from;
1114 user_data_to = map_to;
1116 map_from = cluster_start;
1117 map_to = cluster_end;
1121 * If we haven't allocated the new page yet, we
1122 * shouldn't be writing it out without copying user
1123 * data. This is likely a math error from the caller.
1127 map_from = cluster_start;
1128 map_to = cluster_end;
1130 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1131 cluster_start, cluster_end, new);
1139 * Parts of newly allocated pages need to be zero'd.
1141 * Above, we have also rewritten 'to' and 'from' - as far as
1142 * the rest of the function is concerned, the entire cluster
1143 * range inside of a page needs to be written.
1145 * We can skip this if the page is up to date - it's already
1146 * been zero'd from being read in as a hole.
1148 if (new && !PageUptodate(page))
1149 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1150 cpos, user_data_from, user_data_to);
1152 flush_dcache_page(page);
1159 * This function will only grab one clusters worth of pages.
1161 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1162 struct ocfs2_write_ctxt *wc,
1163 u32 cpos, loff_t user_pos, int new,
1164 struct page *mmap_page)
1167 unsigned long start, target_index, index;
1168 struct inode *inode = mapping->host;
1170 target_index = user_pos >> PAGE_CACHE_SHIFT;
1173 * Figure out how many pages we'll be manipulating here. For
1174 * non allocating write, we just change the one
1175 * page. Otherwise, we'll need a whole clusters worth.
1178 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1179 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1181 wc->w_num_pages = 1;
1182 start = target_index;
1185 for(i = 0; i < wc->w_num_pages; i++) {
1188 if (index == target_index && mmap_page) {
1190 * ocfs2_pagemkwrite() is a little different
1191 * and wants us to directly use the page
1194 lock_page(mmap_page);
1196 if (mmap_page->mapping != mapping) {
1197 unlock_page(mmap_page);
1199 * Sanity check - the locking in
1200 * ocfs2_pagemkwrite() should ensure
1201 * that this code doesn't trigger.
1208 page_cache_get(mmap_page);
1209 wc->w_pages[i] = mmap_page;
1211 wc->w_pages[i] = find_or_create_page(mapping, index,
1213 if (!wc->w_pages[i]) {
1220 if (index == target_index)
1221 wc->w_target_page = wc->w_pages[i];
1228 * Prepare a single cluster for write one cluster into the file.
1230 static int ocfs2_write_cluster(struct address_space *mapping,
1231 u32 phys, unsigned int unwritten,
1232 struct ocfs2_alloc_context *data_ac,
1233 struct ocfs2_alloc_context *meta_ac,
1234 struct ocfs2_write_ctxt *wc, u32 cpos,
1235 loff_t user_pos, unsigned user_len)
1237 int ret, i, new, should_zero = 0;
1238 u64 v_blkno, p_blkno;
1239 struct inode *inode = mapping->host;
1240 struct ocfs2_extent_tree et;
1242 new = phys == 0 ? 1 : 0;
1243 if (new || unwritten)
1250 * This is safe to call with the page locks - it won't take
1251 * any additional semaphores or cluster locks.
1254 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1255 &tmp_pos, 1, 0, wc->w_di_bh,
1256 wc->w_handle, data_ac,
1259 * This shouldn't happen because we must have already
1260 * calculated the correct meta data allocation required. The
1261 * internal tree allocation code should know how to increase
1262 * transaction credits itself.
1264 * If need be, we could handle -EAGAIN for a
1265 * RESTART_TRANS here.
1267 mlog_bug_on_msg(ret == -EAGAIN,
1268 "Inode %llu: EAGAIN return during allocation.\n",
1269 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1274 } else if (unwritten) {
1275 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1276 ret = ocfs2_mark_extent_written(inode, &et,
1277 wc->w_handle, cpos, 1, phys,
1278 meta_ac, &wc->w_dealloc);
1286 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1288 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1291 * The only reason this should fail is due to an inability to
1292 * find the extent added.
1294 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1297 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1298 "at logical block %llu",
1299 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1300 (unsigned long long)v_blkno);
1304 BUG_ON(p_blkno == 0);
1306 for(i = 0; i < wc->w_num_pages; i++) {
1309 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1310 wc->w_pages[i], cpos,
1321 * We only have cleanup to do in case of allocating write.
1324 ocfs2_write_failure(inode, wc, user_pos, user_len);
1331 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1332 struct ocfs2_alloc_context *data_ac,
1333 struct ocfs2_alloc_context *meta_ac,
1334 struct ocfs2_write_ctxt *wc,
1335 loff_t pos, unsigned len)
1339 unsigned int local_len = len;
1340 struct ocfs2_write_cluster_desc *desc;
1341 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1343 for (i = 0; i < wc->w_clen; i++) {
1344 desc = &wc->w_desc[i];
1347 * We have to make sure that the total write passed in
1348 * doesn't extend past a single cluster.
1351 cluster_off = pos & (osb->s_clustersize - 1);
1352 if ((cluster_off + local_len) > osb->s_clustersize)
1353 local_len = osb->s_clustersize - cluster_off;
1355 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1356 desc->c_unwritten, data_ac, meta_ac,
1357 wc, desc->c_cpos, pos, local_len);
1373 * ocfs2_write_end() wants to know which parts of the target page it
1374 * should complete the write on. It's easiest to compute them ahead of
1375 * time when a more complete view of the write is available.
1377 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1378 struct ocfs2_write_ctxt *wc,
1379 loff_t pos, unsigned len, int alloc)
1381 struct ocfs2_write_cluster_desc *desc;
1383 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1384 wc->w_target_to = wc->w_target_from + len;
1390 * Allocating write - we may have different boundaries based
1391 * on page size and cluster size.
1393 * NOTE: We can no longer compute one value from the other as
1394 * the actual write length and user provided length may be
1398 if (wc->w_large_pages) {
1400 * We only care about the 1st and last cluster within
1401 * our range and whether they should be zero'd or not. Either
1402 * value may be extended out to the start/end of a
1403 * newly allocated cluster.
1405 desc = &wc->w_desc[0];
1406 if (ocfs2_should_zero_cluster(desc))
1407 ocfs2_figure_cluster_boundaries(osb,
1412 desc = &wc->w_desc[wc->w_clen - 1];
1413 if (ocfs2_should_zero_cluster(desc))
1414 ocfs2_figure_cluster_boundaries(osb,
1419 wc->w_target_from = 0;
1420 wc->w_target_to = PAGE_CACHE_SIZE;
1425 * Populate each single-cluster write descriptor in the write context
1426 * with information about the i/o to be done.
1428 * Returns the number of clusters that will have to be allocated, as
1429 * well as a worst case estimate of the number of extent records that
1430 * would have to be created during a write to an unwritten region.
1432 static int ocfs2_populate_write_desc(struct inode *inode,
1433 struct ocfs2_write_ctxt *wc,
1434 unsigned int *clusters_to_alloc,
1435 unsigned int *extents_to_split)
1438 struct ocfs2_write_cluster_desc *desc;
1439 unsigned int num_clusters = 0;
1440 unsigned int ext_flags = 0;
1444 *clusters_to_alloc = 0;
1445 *extents_to_split = 0;
1447 for (i = 0; i < wc->w_clen; i++) {
1448 desc = &wc->w_desc[i];
1449 desc->c_cpos = wc->w_cpos + i;
1451 if (num_clusters == 0) {
1453 * Need to look up the next extent record.
1455 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1456 &num_clusters, &ext_flags);
1463 * Assume worst case - that we're writing in
1464 * the middle of the extent.
1466 * We can assume that the write proceeds from
1467 * left to right, in which case the extent
1468 * insert code is smart enough to coalesce the
1469 * next splits into the previous records created.
1471 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1472 *extents_to_split = *extents_to_split + 2;
1475 * Only increment phys if it doesn't describe
1481 desc->c_phys = phys;
1484 *clusters_to_alloc = *clusters_to_alloc + 1;
1486 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1487 desc->c_unwritten = 1;
1497 static int ocfs2_write_begin_inline(struct address_space *mapping,
1498 struct inode *inode,
1499 struct ocfs2_write_ctxt *wc)
1502 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1505 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1507 page = find_or_create_page(mapping, 0, GFP_NOFS);
1514 * If we don't set w_num_pages then this page won't get unlocked
1515 * and freed on cleanup of the write context.
1517 wc->w_pages[0] = wc->w_target_page = page;
1518 wc->w_num_pages = 1;
1520 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1521 if (IS_ERR(handle)) {
1522 ret = PTR_ERR(handle);
1527 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1528 OCFS2_JOURNAL_ACCESS_WRITE);
1530 ocfs2_commit_trans(osb, handle);
1536 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1537 ocfs2_set_inode_data_inline(inode, di);
1539 if (!PageUptodate(page)) {
1540 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1542 ocfs2_commit_trans(osb, handle);
1548 wc->w_handle = handle;
1553 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1555 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1557 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1562 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1563 struct inode *inode, loff_t pos,
1564 unsigned len, struct page *mmap_page,
1565 struct ocfs2_write_ctxt *wc)
1567 int ret, written = 0;
1568 loff_t end = pos + len;
1569 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1571 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1572 (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos,
1573 oi->ip_dyn_features);
1576 * Handle inodes which already have inline data 1st.
1578 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1579 if (mmap_page == NULL &&
1580 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1581 goto do_inline_write;
1584 * The write won't fit - we have to give this inode an
1585 * inline extent list now.
1587 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1594 * Check whether the inode can accept inline data.
1596 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1600 * Check whether the write can fit.
1602 if (mmap_page || end > ocfs2_max_inline_data(inode->i_sb))
1606 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1613 * This signals to the caller that the data can be written
1618 return written ? written : ret;
1622 * This function only does anything for file systems which can't
1623 * handle sparse files.
1625 * What we want to do here is fill in any hole between the current end
1626 * of allocation and the end of our write. That way the rest of the
1627 * write path can treat it as an non-allocating write, which has no
1628 * special case code for sparse/nonsparse files.
1630 static int ocfs2_expand_nonsparse_inode(struct inode *inode, loff_t pos,
1632 struct ocfs2_write_ctxt *wc)
1635 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1636 loff_t newsize = pos + len;
1638 if (ocfs2_sparse_alloc(osb))
1641 if (newsize <= i_size_read(inode))
1644 ret = ocfs2_extend_no_holes(inode, newsize, newsize - len);
1651 int ocfs2_write_begin_nolock(struct address_space *mapping,
1652 loff_t pos, unsigned len, unsigned flags,
1653 struct page **pagep, void **fsdata,
1654 struct buffer_head *di_bh, struct page *mmap_page)
1656 int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
1657 unsigned int clusters_to_alloc, extents_to_split;
1658 struct ocfs2_write_ctxt *wc;
1659 struct inode *inode = mapping->host;
1660 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1661 struct ocfs2_dinode *di;
1662 struct ocfs2_alloc_context *data_ac = NULL;
1663 struct ocfs2_alloc_context *meta_ac = NULL;
1665 struct ocfs2_extent_tree et;
1667 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1673 if (ocfs2_supports_inline_data(osb)) {
1674 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1686 ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc);
1692 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1699 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1702 * We set w_target_from, w_target_to here so that
1703 * ocfs2_write_end() knows which range in the target page to
1704 * write out. An allocation requires that we write the entire
1707 if (clusters_to_alloc || extents_to_split) {
1709 * XXX: We are stretching the limits of
1710 * ocfs2_lock_allocators(). It greatly over-estimates
1711 * the work to be done.
1713 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1714 " clusters_to_add = %u, extents_to_split = %u\n",
1715 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1716 (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
1717 clusters_to_alloc, extents_to_split);
1719 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1720 ret = ocfs2_lock_allocators(inode, &et,
1721 clusters_to_alloc, extents_to_split,
1722 &data_ac, &meta_ac);
1728 credits = ocfs2_calc_extend_credits(inode->i_sb,
1734 ocfs2_set_target_boundaries(osb, wc, pos, len,
1735 clusters_to_alloc + extents_to_split);
1737 handle = ocfs2_start_trans(osb, credits);
1738 if (IS_ERR(handle)) {
1739 ret = PTR_ERR(handle);
1744 wc->w_handle = handle;
1747 * We don't want this to fail in ocfs2_write_end(), so do it
1750 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1751 OCFS2_JOURNAL_ACCESS_WRITE);
1758 * Fill our page array first. That way we've grabbed enough so
1759 * that we can zero and flush if we error after adding the
1762 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
1763 clusters_to_alloc + extents_to_split,
1770 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1778 ocfs2_free_alloc_context(data_ac);
1780 ocfs2_free_alloc_context(meta_ac);
1783 *pagep = wc->w_target_page;
1787 ocfs2_commit_trans(osb, handle);
1790 ocfs2_free_write_ctxt(wc);
1793 ocfs2_free_alloc_context(data_ac);
1795 ocfs2_free_alloc_context(meta_ac);
1799 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1800 loff_t pos, unsigned len, unsigned flags,
1801 struct page **pagep, void **fsdata)
1804 struct buffer_head *di_bh = NULL;
1805 struct inode *inode = mapping->host;
1807 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1814 * Take alloc sem here to prevent concurrent lookups. That way
1815 * the mapping, zeroing and tree manipulation within
1816 * ocfs2_write() will be safe against ->readpage(). This
1817 * should also serve to lock out allocation from a shared
1820 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1822 ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
1823 fsdata, di_bh, NULL);
1834 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1837 ocfs2_inode_unlock(inode, 1);
1842 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1843 unsigned len, unsigned *copied,
1844 struct ocfs2_dinode *di,
1845 struct ocfs2_write_ctxt *wc)
1849 if (unlikely(*copied < len)) {
1850 if (!PageUptodate(wc->w_target_page)) {
1856 kaddr = kmap_atomic(wc->w_target_page, KM_USER0);
1857 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1858 kunmap_atomic(kaddr, KM_USER0);
1860 mlog(0, "Data written to inode at offset %llu. "
1861 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1862 (unsigned long long)pos, *copied,
1863 le16_to_cpu(di->id2.i_data.id_count),
1864 le16_to_cpu(di->i_dyn_features));
1867 int ocfs2_write_end_nolock(struct address_space *mapping,
1868 loff_t pos, unsigned len, unsigned copied,
1869 struct page *page, void *fsdata)
1872 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1873 struct inode *inode = mapping->host;
1874 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1875 struct ocfs2_write_ctxt *wc = fsdata;
1876 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1877 handle_t *handle = wc->w_handle;
1878 struct page *tmppage;
1880 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1881 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1882 goto out_write_size;
1885 if (unlikely(copied < len)) {
1886 if (!PageUptodate(wc->w_target_page))
1889 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1892 flush_dcache_page(wc->w_target_page);
1894 for(i = 0; i < wc->w_num_pages; i++) {
1895 tmppage = wc->w_pages[i];
1897 if (tmppage == wc->w_target_page) {
1898 from = wc->w_target_from;
1899 to = wc->w_target_to;
1901 BUG_ON(from > PAGE_CACHE_SIZE ||
1902 to > PAGE_CACHE_SIZE ||
1906 * Pages adjacent to the target (if any) imply
1907 * a hole-filling write in which case we want
1908 * to flush their entire range.
1911 to = PAGE_CACHE_SIZE;
1914 if (page_has_buffers(tmppage)) {
1915 if (ocfs2_should_order_data(inode)) {
1916 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1917 #ifdef CONFIG_OCFS2_COMPAT_JBD
1918 walk_page_buffers(wc->w_handle,
1919 page_buffers(tmppage),
1921 ocfs2_journal_dirty_data);
1924 block_commit_write(tmppage, from, to);
1930 if (pos > inode->i_size) {
1931 i_size_write(inode, pos);
1932 mark_inode_dirty(inode);
1934 inode->i_blocks = ocfs2_inode_sector_count(inode);
1935 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1936 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1937 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1938 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1939 ocfs2_journal_dirty(handle, wc->w_di_bh);
1941 ocfs2_commit_trans(osb, handle);
1943 ocfs2_run_deallocs(osb, &wc->w_dealloc);
1945 ocfs2_free_write_ctxt(wc);
1950 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
1951 loff_t pos, unsigned len, unsigned copied,
1952 struct page *page, void *fsdata)
1955 struct inode *inode = mapping->host;
1957 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
1959 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1960 ocfs2_inode_unlock(inode, 1);
1965 const struct address_space_operations ocfs2_aops = {
1966 .readpage = ocfs2_readpage,
1967 .readpages = ocfs2_readpages,
1968 .writepage = ocfs2_writepage,
1969 .write_begin = ocfs2_write_begin,
1970 .write_end = ocfs2_write_end,
1972 .sync_page = block_sync_page,
1973 .direct_IO = ocfs2_direct_IO,
1974 .invalidatepage = ocfs2_invalidatepage,
1975 .releasepage = ocfs2_releasepage,
1976 .migratepage = buffer_migrate_page,