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>
30 #include <linux/quotaops.h>
32 #define MLOG_MASK_PREFIX ML_FILE_IO
33 #include <cluster/masklog.h>
40 #include "extent_map.h"
48 #include "buffer_head_io.h"
50 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
51 struct buffer_head *bh_result, int create)
55 struct ocfs2_dinode *fe = NULL;
56 struct buffer_head *bh = NULL;
57 struct buffer_head *buffer_cache_bh = NULL;
58 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
61 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
62 (unsigned long long)iblock, bh_result, create);
64 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
66 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
67 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
68 (unsigned long long)iblock);
72 status = ocfs2_read_inode_block(inode, &bh);
77 fe = (struct ocfs2_dinode *) bh->b_data;
79 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
80 le32_to_cpu(fe->i_clusters))) {
81 mlog(ML_ERROR, "block offset is outside the allocated size: "
82 "%llu\n", (unsigned long long)iblock);
86 /* We don't use the page cache to create symlink data, so if
87 * need be, copy it over from the buffer cache. */
88 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
89 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
91 buffer_cache_bh = sb_getblk(osb->sb, blkno);
92 if (!buffer_cache_bh) {
93 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
97 /* we haven't locked out transactions, so a commit
98 * could've happened. Since we've got a reference on
99 * the bh, even if it commits while we're doing the
100 * copy, the data is still good. */
101 if (buffer_jbd(buffer_cache_bh)
102 && ocfs2_inode_is_new(inode)) {
103 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
105 mlog(ML_ERROR, "couldn't kmap!\n");
108 memcpy(kaddr + (bh_result->b_size * iblock),
109 buffer_cache_bh->b_data,
111 kunmap_atomic(kaddr, KM_USER0);
112 set_buffer_uptodate(bh_result);
114 brelse(buffer_cache_bh);
117 map_bh(bh_result, inode->i_sb,
118 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
129 static int ocfs2_get_block(struct inode *inode, sector_t iblock,
130 struct buffer_head *bh_result, int create)
133 unsigned int ext_flags;
134 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
135 u64 p_blkno, count, past_eof;
136 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
138 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
139 (unsigned long long)iblock, bh_result, create);
141 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
142 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
143 inode, inode->i_ino);
145 if (S_ISLNK(inode->i_mode)) {
146 /* this always does I/O for some reason. */
147 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
151 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
154 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
155 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
156 (unsigned long long)p_blkno);
160 if (max_blocks < count)
164 * ocfs2 never allocates in this function - the only time we
165 * need to use BH_New is when we're extending i_size on a file
166 * system which doesn't support holes, in which case BH_New
167 * allows block_prepare_write() to zero.
169 * If we see this on a sparse file system, then a truncate has
170 * raced us and removed the cluster. In this case, we clear
171 * the buffers dirty and uptodate bits and let the buffer code
172 * ignore it as a hole.
174 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
175 clear_buffer_dirty(bh_result);
176 clear_buffer_uptodate(bh_result);
180 /* Treat the unwritten extent as a hole for zeroing purposes. */
181 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
182 map_bh(bh_result, inode->i_sb, p_blkno);
184 bh_result->b_size = count << inode->i_blkbits;
186 if (!ocfs2_sparse_alloc(osb)) {
190 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
191 (unsigned long long)iblock,
192 (unsigned long long)p_blkno,
193 (unsigned long long)OCFS2_I(inode)->ip_blkno);
194 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
199 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
200 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
201 (unsigned long long)past_eof);
203 if (create && (iblock >= past_eof))
204 set_buffer_new(bh_result);
215 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
216 struct buffer_head *di_bh)
220 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
222 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
223 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
224 (unsigned long long)OCFS2_I(inode)->ip_blkno);
228 size = i_size_read(inode);
230 if (size > PAGE_CACHE_SIZE ||
231 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
232 ocfs2_error(inode->i_sb,
233 "Inode %llu has with inline data has bad size: %Lu",
234 (unsigned long long)OCFS2_I(inode)->ip_blkno,
235 (unsigned long long)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;
257 BUG_ON(!PageLocked(page));
258 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
260 ret = ocfs2_read_inode_block(inode, &di_bh);
266 ret = ocfs2_read_inline_data(inode, page, di_bh);
274 static int ocfs2_readpage(struct file *file, struct page *page)
276 struct inode *inode = page->mapping->host;
277 struct ocfs2_inode_info *oi = OCFS2_I(inode);
278 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
281 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
283 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
285 if (ret == AOP_TRUNCATED_PAGE)
291 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
292 ret = AOP_TRUNCATED_PAGE;
293 goto out_inode_unlock;
297 * i_size might have just been updated as we grabed the meta lock. We
298 * might now be discovering a truncate that hit on another node.
299 * block_read_full_page->get_block freaks out if it is asked to read
300 * beyond the end of a file, so we check here. Callers
301 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
302 * and notice that the page they just read isn't needed.
304 * XXX sys_readahead() seems to get that wrong?
306 if (start >= i_size_read(inode)) {
307 zero_user(page, 0, PAGE_SIZE);
308 SetPageUptodate(page);
313 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
314 ret = ocfs2_readpage_inline(inode, page);
316 ret = block_read_full_page(page, ocfs2_get_block);
320 up_read(&OCFS2_I(inode)->ip_alloc_sem);
322 ocfs2_inode_unlock(inode, 0);
331 * This is used only for read-ahead. Failures or difficult to handle
332 * situations are safe to ignore.
334 * Right now, we don't bother with BH_Boundary - in-inode extent lists
335 * are quite large (243 extents on 4k blocks), so most inodes don't
336 * grow out to a tree. If need be, detecting boundary extents could
337 * trivially be added in a future version of ocfs2_get_block().
339 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
340 struct list_head *pages, unsigned nr_pages)
343 struct inode *inode = mapping->host;
344 struct ocfs2_inode_info *oi = OCFS2_I(inode);
349 * Use the nonblocking flag for the dlm code to avoid page
350 * lock inversion, but don't bother with retrying.
352 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
356 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
357 ocfs2_inode_unlock(inode, 0);
362 * Don't bother with inline-data. There isn't anything
363 * to read-ahead in that case anyway...
365 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
369 * Check whether a remote node truncated this file - we just
370 * drop out in that case as it's not worth handling here.
372 last = list_entry(pages->prev, struct page, lru);
373 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
374 if (start >= i_size_read(inode))
377 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
380 up_read(&oi->ip_alloc_sem);
381 ocfs2_inode_unlock(inode, 0);
386 /* Note: Because we don't support holes, our allocation has
387 * already happened (allocation writes zeros to the file data)
388 * so we don't have to worry about ordered writes in
391 * ->writepage is called during the process of invalidating the page cache
392 * during blocked lock processing. It can't block on any cluster locks
393 * to during block mapping. It's relying on the fact that the block
394 * mapping can't have disappeared under the dirty pages that it is
395 * being asked to write back.
397 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
401 mlog_entry("(0x%p)\n", page);
403 ret = block_write_full_page(page, ocfs2_get_block, wbc);
411 * This is called from ocfs2_write_zero_page() which has handled it's
412 * own cluster locking and has ensured allocation exists for those
413 * blocks to be written.
415 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
416 unsigned from, unsigned to)
420 ret = block_prepare_write(page, from, to, ocfs2_get_block);
425 /* Taken from ext3. We don't necessarily need the full blown
426 * functionality yet, but IMHO it's better to cut and paste the whole
427 * thing so we can avoid introducing our own bugs (and easily pick up
428 * their fixes when they happen) --Mark */
429 int walk_page_buffers( handle_t *handle,
430 struct buffer_head *head,
434 int (*fn)( handle_t *handle,
435 struct buffer_head *bh))
437 struct buffer_head *bh;
438 unsigned block_start, block_end;
439 unsigned blocksize = head->b_size;
441 struct buffer_head *next;
443 for ( bh = head, block_start = 0;
444 ret == 0 && (bh != head || !block_start);
445 block_start = block_end, bh = next)
447 next = bh->b_this_page;
448 block_end = block_start + blocksize;
449 if (block_end <= from || block_start >= to) {
450 if (partial && !buffer_uptodate(bh))
454 err = (*fn)(handle, bh);
461 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
466 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
470 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
471 if (IS_ERR(handle)) {
477 if (ocfs2_should_order_data(inode)) {
478 ret = ocfs2_jbd2_file_inode(handle, inode);
485 ocfs2_commit_trans(osb, handle);
486 handle = ERR_PTR(ret);
491 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
496 struct inode *inode = mapping->host;
498 mlog_entry("(block = %llu)\n", (unsigned long long)block);
500 /* We don't need to lock journal system files, since they aren't
501 * accessed concurrently from multiple nodes.
503 if (!INODE_JOURNAL(inode)) {
504 err = ocfs2_inode_lock(inode, NULL, 0);
510 down_read(&OCFS2_I(inode)->ip_alloc_sem);
513 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
514 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
517 if (!INODE_JOURNAL(inode)) {
518 up_read(&OCFS2_I(inode)->ip_alloc_sem);
519 ocfs2_inode_unlock(inode, 0);
523 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
524 (unsigned long long)block);
530 status = err ? 0 : p_blkno;
532 mlog_exit((int)status);
538 * TODO: Make this into a generic get_blocks function.
540 * From do_direct_io in direct-io.c:
541 * "So what we do is to permit the ->get_blocks function to populate
542 * bh.b_size with the size of IO which is permitted at this offset and
545 * This function is called directly from get_more_blocks in direct-io.c.
547 * called like this: dio->get_blocks(dio->inode, fs_startblk,
548 * fs_count, map_bh, dio->rw == WRITE);
550 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
551 struct buffer_head *bh_result, int create)
554 u64 p_blkno, inode_blocks, contig_blocks;
555 unsigned int ext_flags;
556 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
557 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
559 /* This function won't even be called if the request isn't all
560 * nicely aligned and of the right size, so there's no need
561 * for us to check any of that. */
563 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
566 * Any write past EOF is not allowed because we'd be extending.
568 if (create && (iblock + max_blocks) > inode_blocks) {
573 /* This figures out the size of the next contiguous block, and
574 * our logical offset */
575 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
576 &contig_blocks, &ext_flags);
578 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
579 (unsigned long long)iblock);
584 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno && create) {
585 ocfs2_error(inode->i_sb,
586 "Inode %llu has a hole at block %llu\n",
587 (unsigned long long)OCFS2_I(inode)->ip_blkno,
588 (unsigned long long)iblock);
594 * get_more_blocks() expects us to describe a hole by clearing
595 * the mapped bit on bh_result().
597 * Consider an unwritten extent as a hole.
599 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
600 map_bh(bh_result, inode->i_sb, p_blkno);
603 * ocfs2_prepare_inode_for_write() should have caught
604 * the case where we'd be filling a hole and triggered
605 * a buffered write instead.
613 clear_buffer_mapped(bh_result);
616 /* make sure we don't map more than max_blocks blocks here as
617 that's all the kernel will handle at this point. */
618 if (max_blocks < contig_blocks)
619 contig_blocks = max_blocks;
620 bh_result->b_size = contig_blocks << blocksize_bits;
626 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
627 * particularly interested in the aio/dio case. Like the core uses
628 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
629 * truncation on another.
631 static void ocfs2_dio_end_io(struct kiocb *iocb,
636 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
639 /* this io's submitter should not have unlocked this before we could */
640 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
642 ocfs2_iocb_clear_rw_locked(iocb);
644 level = ocfs2_iocb_rw_locked_level(iocb);
646 up_read(&inode->i_alloc_sem);
647 ocfs2_rw_unlock(inode, level);
651 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
652 * from ext3. PageChecked() bits have been removed as OCFS2 does not
653 * do journalled data.
655 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
657 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
659 jbd2_journal_invalidatepage(journal, page, offset);
662 static int ocfs2_releasepage(struct page *page, gfp_t wait)
664 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
666 if (!page_has_buffers(page))
668 return jbd2_journal_try_to_free_buffers(journal, page, wait);
671 static ssize_t ocfs2_direct_IO(int rw,
673 const struct iovec *iov,
675 unsigned long nr_segs)
677 struct file *file = iocb->ki_filp;
678 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
684 * Fallback to buffered I/O if we see an inode without
687 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
690 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
691 inode->i_sb->s_bdev, iov, offset,
693 ocfs2_direct_IO_get_blocks,
700 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
705 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
707 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
710 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
712 cluster_start = cpos % cpp;
713 cluster_start = cluster_start << osb->s_clustersize_bits;
715 cluster_end = cluster_start + osb->s_clustersize;
718 BUG_ON(cluster_start > PAGE_SIZE);
719 BUG_ON(cluster_end > PAGE_SIZE);
722 *start = cluster_start;
728 * 'from' and 'to' are the region in the page to avoid zeroing.
730 * If pagesize > clustersize, this function will avoid zeroing outside
731 * of the cluster boundary.
733 * from == to == 0 is code for "zero the entire cluster region"
735 static void ocfs2_clear_page_regions(struct page *page,
736 struct ocfs2_super *osb, u32 cpos,
737 unsigned from, unsigned to)
740 unsigned int cluster_start, cluster_end;
742 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
744 kaddr = kmap_atomic(page, KM_USER0);
747 if (from > cluster_start)
748 memset(kaddr + cluster_start, 0, from - cluster_start);
749 if (to < cluster_end)
750 memset(kaddr + to, 0, cluster_end - to);
752 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
755 kunmap_atomic(kaddr, KM_USER0);
759 * Nonsparse file systems fully allocate before we get to the write
760 * code. This prevents ocfs2_write() from tagging the write as an
761 * allocating one, which means ocfs2_map_page_blocks() might try to
762 * read-in the blocks at the tail of our file. Avoid reading them by
763 * testing i_size against each block offset.
765 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
766 unsigned int block_start)
768 u64 offset = page_offset(page) + block_start;
770 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
773 if (i_size_read(inode) > offset)
780 * Some of this taken from block_prepare_write(). We already have our
781 * mapping by now though, and the entire write will be allocating or
782 * it won't, so not much need to use BH_New.
784 * This will also skip zeroing, which is handled externally.
786 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
787 struct inode *inode, unsigned int from,
788 unsigned int to, int new)
791 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
792 unsigned int block_end, block_start;
793 unsigned int bsize = 1 << inode->i_blkbits;
795 if (!page_has_buffers(page))
796 create_empty_buffers(page, bsize, 0);
798 head = page_buffers(page);
799 for (bh = head, block_start = 0; bh != head || !block_start;
800 bh = bh->b_this_page, block_start += bsize) {
801 block_end = block_start + bsize;
803 clear_buffer_new(bh);
806 * Ignore blocks outside of our i/o range -
807 * they may belong to unallocated clusters.
809 if (block_start >= to || block_end <= from) {
810 if (PageUptodate(page))
811 set_buffer_uptodate(bh);
816 * For an allocating write with cluster size >= page
817 * size, we always write the entire page.
822 if (!buffer_mapped(bh)) {
823 map_bh(bh, inode->i_sb, *p_blkno);
824 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
827 if (PageUptodate(page)) {
828 if (!buffer_uptodate(bh))
829 set_buffer_uptodate(bh);
830 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
832 ocfs2_should_read_blk(inode, page, block_start) &&
833 (block_start < from || block_end > to)) {
834 ll_rw_block(READ, 1, &bh);
838 *p_blkno = *p_blkno + 1;
842 * If we issued read requests - let them complete.
844 while(wait_bh > wait) {
845 wait_on_buffer(*--wait_bh);
846 if (!buffer_uptodate(*wait_bh))
850 if (ret == 0 || !new)
854 * If we get -EIO above, zero out any newly allocated blocks
855 * to avoid exposing stale data.
860 block_end = block_start + bsize;
861 if (block_end <= from)
863 if (block_start >= to)
866 zero_user(page, block_start, bh->b_size);
867 set_buffer_uptodate(bh);
868 mark_buffer_dirty(bh);
871 block_start = block_end;
872 bh = bh->b_this_page;
873 } while (bh != head);
878 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
879 #define OCFS2_MAX_CTXT_PAGES 1
881 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
884 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
887 * Describe the state of a single cluster to be written to.
889 struct ocfs2_write_cluster_desc {
893 * Give this a unique field because c_phys eventually gets
897 unsigned c_unwritten;
900 static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
902 return d->c_new || d->c_unwritten;
905 struct ocfs2_write_ctxt {
906 /* Logical cluster position / len of write */
910 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
913 * This is true if page_size > cluster_size.
915 * It triggers a set of special cases during write which might
916 * have to deal with allocating writes to partial pages.
918 unsigned int w_large_pages;
921 * Pages involved in this write.
923 * w_target_page is the page being written to by the user.
925 * w_pages is an array of pages which always contains
926 * w_target_page, and in the case of an allocating write with
927 * page_size < cluster size, it will contain zero'd and mapped
928 * pages adjacent to w_target_page which need to be written
929 * out in so that future reads from that region will get
932 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
933 unsigned int w_num_pages;
934 struct page *w_target_page;
937 * ocfs2_write_end() uses this to know what the real range to
938 * write in the target should be.
940 unsigned int w_target_from;
941 unsigned int w_target_to;
944 * We could use journal_current_handle() but this is cleaner,
949 struct buffer_head *w_di_bh;
951 struct ocfs2_cached_dealloc_ctxt w_dealloc;
954 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
958 for(i = 0; i < num_pages; i++) {
960 unlock_page(pages[i]);
961 mark_page_accessed(pages[i]);
962 page_cache_release(pages[i]);
967 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
969 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
975 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
976 struct ocfs2_super *osb, loff_t pos,
977 unsigned len, struct buffer_head *di_bh)
980 struct ocfs2_write_ctxt *wc;
982 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
986 wc->w_cpos = pos >> osb->s_clustersize_bits;
987 cend = (pos + len - 1) >> osb->s_clustersize_bits;
988 wc->w_clen = cend - wc->w_cpos + 1;
992 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
993 wc->w_large_pages = 1;
995 wc->w_large_pages = 0;
997 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1005 * If a page has any new buffers, zero them out here, and mark them uptodate
1006 * and dirty so they'll be written out (in order to prevent uninitialised
1007 * block data from leaking). And clear the new bit.
1009 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1011 unsigned int block_start, block_end;
1012 struct buffer_head *head, *bh;
1014 BUG_ON(!PageLocked(page));
1015 if (!page_has_buffers(page))
1018 bh = head = page_buffers(page);
1021 block_end = block_start + bh->b_size;
1023 if (buffer_new(bh)) {
1024 if (block_end > from && block_start < to) {
1025 if (!PageUptodate(page)) {
1026 unsigned start, end;
1028 start = max(from, block_start);
1029 end = min(to, block_end);
1031 zero_user_segment(page, start, end);
1032 set_buffer_uptodate(bh);
1035 clear_buffer_new(bh);
1036 mark_buffer_dirty(bh);
1040 block_start = block_end;
1041 bh = bh->b_this_page;
1042 } while (bh != head);
1046 * Only called when we have a failure during allocating write to write
1047 * zero's to the newly allocated region.
1049 static void ocfs2_write_failure(struct inode *inode,
1050 struct ocfs2_write_ctxt *wc,
1051 loff_t user_pos, unsigned user_len)
1054 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1055 to = user_pos + user_len;
1056 struct page *tmppage;
1058 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1060 for(i = 0; i < wc->w_num_pages; i++) {
1061 tmppage = wc->w_pages[i];
1063 if (page_has_buffers(tmppage)) {
1064 if (ocfs2_should_order_data(inode))
1065 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1067 block_commit_write(tmppage, from, to);
1072 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1073 struct ocfs2_write_ctxt *wc,
1074 struct page *page, u32 cpos,
1075 loff_t user_pos, unsigned user_len,
1079 unsigned int map_from = 0, map_to = 0;
1080 unsigned int cluster_start, cluster_end;
1081 unsigned int user_data_from = 0, user_data_to = 0;
1083 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1084 &cluster_start, &cluster_end);
1086 if (page == wc->w_target_page) {
1087 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1088 map_to = map_from + user_len;
1091 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1092 cluster_start, cluster_end,
1095 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1096 map_from, map_to, new);
1102 user_data_from = map_from;
1103 user_data_to = map_to;
1105 map_from = cluster_start;
1106 map_to = cluster_end;
1110 * If we haven't allocated the new page yet, we
1111 * shouldn't be writing it out without copying user
1112 * data. This is likely a math error from the caller.
1116 map_from = cluster_start;
1117 map_to = cluster_end;
1119 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1120 cluster_start, cluster_end, new);
1128 * Parts of newly allocated pages need to be zero'd.
1130 * Above, we have also rewritten 'to' and 'from' - as far as
1131 * the rest of the function is concerned, the entire cluster
1132 * range inside of a page needs to be written.
1134 * We can skip this if the page is up to date - it's already
1135 * been zero'd from being read in as a hole.
1137 if (new && !PageUptodate(page))
1138 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1139 cpos, user_data_from, user_data_to);
1141 flush_dcache_page(page);
1148 * This function will only grab one clusters worth of pages.
1150 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1151 struct ocfs2_write_ctxt *wc,
1152 u32 cpos, loff_t user_pos, int new,
1153 struct page *mmap_page)
1156 unsigned long start, target_index, index;
1157 struct inode *inode = mapping->host;
1159 target_index = user_pos >> PAGE_CACHE_SHIFT;
1162 * Figure out how many pages we'll be manipulating here. For
1163 * non allocating write, we just change the one
1164 * page. Otherwise, we'll need a whole clusters worth.
1167 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1168 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1170 wc->w_num_pages = 1;
1171 start = target_index;
1174 for(i = 0; i < wc->w_num_pages; i++) {
1177 if (index == target_index && mmap_page) {
1179 * ocfs2_pagemkwrite() is a little different
1180 * and wants us to directly use the page
1183 lock_page(mmap_page);
1185 if (mmap_page->mapping != mapping) {
1186 unlock_page(mmap_page);
1188 * Sanity check - the locking in
1189 * ocfs2_pagemkwrite() should ensure
1190 * that this code doesn't trigger.
1197 page_cache_get(mmap_page);
1198 wc->w_pages[i] = mmap_page;
1200 wc->w_pages[i] = find_or_create_page(mapping, index,
1202 if (!wc->w_pages[i]) {
1209 if (index == target_index)
1210 wc->w_target_page = wc->w_pages[i];
1217 * Prepare a single cluster for write one cluster into the file.
1219 static int ocfs2_write_cluster(struct address_space *mapping,
1220 u32 phys, unsigned int unwritten,
1221 struct ocfs2_alloc_context *data_ac,
1222 struct ocfs2_alloc_context *meta_ac,
1223 struct ocfs2_write_ctxt *wc, u32 cpos,
1224 loff_t user_pos, unsigned user_len)
1226 int ret, i, new, should_zero = 0;
1227 u64 v_blkno, p_blkno;
1228 struct inode *inode = mapping->host;
1229 struct ocfs2_extent_tree et;
1231 new = phys == 0 ? 1 : 0;
1232 if (new || unwritten)
1239 * This is safe to call with the page locks - it won't take
1240 * any additional semaphores or cluster locks.
1243 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1244 &tmp_pos, 1, 0, wc->w_di_bh,
1245 wc->w_handle, data_ac,
1248 * This shouldn't happen because we must have already
1249 * calculated the correct meta data allocation required. The
1250 * internal tree allocation code should know how to increase
1251 * transaction credits itself.
1253 * If need be, we could handle -EAGAIN for a
1254 * RESTART_TRANS here.
1256 mlog_bug_on_msg(ret == -EAGAIN,
1257 "Inode %llu: EAGAIN return during allocation.\n",
1258 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1263 } else if (unwritten) {
1264 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1265 ret = ocfs2_mark_extent_written(inode, &et,
1266 wc->w_handle, cpos, 1, phys,
1267 meta_ac, &wc->w_dealloc);
1275 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1277 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1280 * The only reason this should fail is due to an inability to
1281 * find the extent added.
1283 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1286 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1287 "at logical block %llu",
1288 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1289 (unsigned long long)v_blkno);
1293 BUG_ON(p_blkno == 0);
1295 for(i = 0; i < wc->w_num_pages; i++) {
1298 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1299 wc->w_pages[i], cpos,
1310 * We only have cleanup to do in case of allocating write.
1313 ocfs2_write_failure(inode, wc, user_pos, user_len);
1320 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1321 struct ocfs2_alloc_context *data_ac,
1322 struct ocfs2_alloc_context *meta_ac,
1323 struct ocfs2_write_ctxt *wc,
1324 loff_t pos, unsigned len)
1328 unsigned int local_len = len;
1329 struct ocfs2_write_cluster_desc *desc;
1330 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1332 for (i = 0; i < wc->w_clen; i++) {
1333 desc = &wc->w_desc[i];
1336 * We have to make sure that the total write passed in
1337 * doesn't extend past a single cluster.
1340 cluster_off = pos & (osb->s_clustersize - 1);
1341 if ((cluster_off + local_len) > osb->s_clustersize)
1342 local_len = osb->s_clustersize - cluster_off;
1344 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1345 desc->c_unwritten, data_ac, meta_ac,
1346 wc, desc->c_cpos, pos, local_len);
1362 * ocfs2_write_end() wants to know which parts of the target page it
1363 * should complete the write on. It's easiest to compute them ahead of
1364 * time when a more complete view of the write is available.
1366 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1367 struct ocfs2_write_ctxt *wc,
1368 loff_t pos, unsigned len, int alloc)
1370 struct ocfs2_write_cluster_desc *desc;
1372 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1373 wc->w_target_to = wc->w_target_from + len;
1379 * Allocating write - we may have different boundaries based
1380 * on page size and cluster size.
1382 * NOTE: We can no longer compute one value from the other as
1383 * the actual write length and user provided length may be
1387 if (wc->w_large_pages) {
1389 * We only care about the 1st and last cluster within
1390 * our range and whether they should be zero'd or not. Either
1391 * value may be extended out to the start/end of a
1392 * newly allocated cluster.
1394 desc = &wc->w_desc[0];
1395 if (ocfs2_should_zero_cluster(desc))
1396 ocfs2_figure_cluster_boundaries(osb,
1401 desc = &wc->w_desc[wc->w_clen - 1];
1402 if (ocfs2_should_zero_cluster(desc))
1403 ocfs2_figure_cluster_boundaries(osb,
1408 wc->w_target_from = 0;
1409 wc->w_target_to = PAGE_CACHE_SIZE;
1414 * Populate each single-cluster write descriptor in the write context
1415 * with information about the i/o to be done.
1417 * Returns the number of clusters that will have to be allocated, as
1418 * well as a worst case estimate of the number of extent records that
1419 * would have to be created during a write to an unwritten region.
1421 static int ocfs2_populate_write_desc(struct inode *inode,
1422 struct ocfs2_write_ctxt *wc,
1423 unsigned int *clusters_to_alloc,
1424 unsigned int *extents_to_split)
1427 struct ocfs2_write_cluster_desc *desc;
1428 unsigned int num_clusters = 0;
1429 unsigned int ext_flags = 0;
1433 *clusters_to_alloc = 0;
1434 *extents_to_split = 0;
1436 for (i = 0; i < wc->w_clen; i++) {
1437 desc = &wc->w_desc[i];
1438 desc->c_cpos = wc->w_cpos + i;
1440 if (num_clusters == 0) {
1442 * Need to look up the next extent record.
1444 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1445 &num_clusters, &ext_flags);
1452 * Assume worst case - that we're writing in
1453 * the middle of the extent.
1455 * We can assume that the write proceeds from
1456 * left to right, in which case the extent
1457 * insert code is smart enough to coalesce the
1458 * next splits into the previous records created.
1460 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1461 *extents_to_split = *extents_to_split + 2;
1464 * Only increment phys if it doesn't describe
1470 desc->c_phys = phys;
1473 *clusters_to_alloc = *clusters_to_alloc + 1;
1475 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1476 desc->c_unwritten = 1;
1486 static int ocfs2_write_begin_inline(struct address_space *mapping,
1487 struct inode *inode,
1488 struct ocfs2_write_ctxt *wc)
1491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1494 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1496 page = find_or_create_page(mapping, 0, GFP_NOFS);
1503 * If we don't set w_num_pages then this page won't get unlocked
1504 * and freed on cleanup of the write context.
1506 wc->w_pages[0] = wc->w_target_page = page;
1507 wc->w_num_pages = 1;
1509 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1510 if (IS_ERR(handle)) {
1511 ret = PTR_ERR(handle);
1516 ret = ocfs2_journal_access_di(handle, inode, wc->w_di_bh,
1517 OCFS2_JOURNAL_ACCESS_WRITE);
1519 ocfs2_commit_trans(osb, handle);
1525 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1526 ocfs2_set_inode_data_inline(inode, di);
1528 if (!PageUptodate(page)) {
1529 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1531 ocfs2_commit_trans(osb, handle);
1537 wc->w_handle = handle;
1542 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1544 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1546 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1551 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1552 struct inode *inode, loff_t pos,
1553 unsigned len, struct page *mmap_page,
1554 struct ocfs2_write_ctxt *wc)
1556 int ret, written = 0;
1557 loff_t end = pos + len;
1558 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1559 struct ocfs2_dinode *di = NULL;
1561 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1562 (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos,
1563 oi->ip_dyn_features);
1566 * Handle inodes which already have inline data 1st.
1568 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1569 if (mmap_page == NULL &&
1570 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1571 goto do_inline_write;
1574 * The write won't fit - we have to give this inode an
1575 * inline extent list now.
1577 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1584 * Check whether the inode can accept inline data.
1586 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1590 * Check whether the write can fit.
1592 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1594 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
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;
1657 struct ocfs2_extent_tree et;
1659 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1665 if (ocfs2_supports_inline_data(osb)) {
1666 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1678 ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc);
1684 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1691 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1694 * We set w_target_from, w_target_to here so that
1695 * ocfs2_write_end() knows which range in the target page to
1696 * write out. An allocation requires that we write the entire
1699 if (clusters_to_alloc || extents_to_split) {
1701 * XXX: We are stretching the limits of
1702 * ocfs2_lock_allocators(). It greatly over-estimates
1703 * the work to be done.
1705 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1706 " clusters_to_add = %u, extents_to_split = %u\n",
1707 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1708 (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
1709 clusters_to_alloc, extents_to_split);
1711 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1712 ret = ocfs2_lock_allocators(inode, &et,
1713 clusters_to_alloc, extents_to_split,
1714 &data_ac, &meta_ac);
1720 credits = ocfs2_calc_extend_credits(inode->i_sb,
1726 ocfs2_set_target_boundaries(osb, wc, pos, len,
1727 clusters_to_alloc + extents_to_split);
1729 handle = ocfs2_start_trans(osb, credits);
1730 if (IS_ERR(handle)) {
1731 ret = PTR_ERR(handle);
1736 wc->w_handle = handle;
1738 if (clusters_to_alloc && vfs_dq_alloc_space_nodirty(inode,
1739 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc))) {
1744 * We don't want this to fail in ocfs2_write_end(), so do it
1747 ret = ocfs2_journal_access_di(handle, inode, wc->w_di_bh,
1748 OCFS2_JOURNAL_ACCESS_WRITE);
1755 * Fill our page array first. That way we've grabbed enough so
1756 * that we can zero and flush if we error after adding the
1759 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
1760 clusters_to_alloc + extents_to_split,
1767 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1775 ocfs2_free_alloc_context(data_ac);
1777 ocfs2_free_alloc_context(meta_ac);
1780 *pagep = wc->w_target_page;
1784 if (clusters_to_alloc)
1785 vfs_dq_free_space(inode,
1786 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1788 ocfs2_commit_trans(osb, handle);
1791 ocfs2_free_write_ctxt(wc);
1794 ocfs2_free_alloc_context(data_ac);
1796 ocfs2_free_alloc_context(meta_ac);
1800 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1801 loff_t pos, unsigned len, unsigned flags,
1802 struct page **pagep, void **fsdata)
1805 struct buffer_head *di_bh = NULL;
1806 struct inode *inode = mapping->host;
1808 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1815 * Take alloc sem here to prevent concurrent lookups. That way
1816 * the mapping, zeroing and tree manipulation within
1817 * ocfs2_write() will be safe against ->readpage(). This
1818 * should also serve to lock out allocation from a shared
1821 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1823 ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
1824 fsdata, di_bh, NULL);
1835 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1838 ocfs2_inode_unlock(inode, 1);
1843 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1844 unsigned len, unsigned *copied,
1845 struct ocfs2_dinode *di,
1846 struct ocfs2_write_ctxt *wc)
1850 if (unlikely(*copied < len)) {
1851 if (!PageUptodate(wc->w_target_page)) {
1857 kaddr = kmap_atomic(wc->w_target_page, KM_USER0);
1858 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1859 kunmap_atomic(kaddr, KM_USER0);
1861 mlog(0, "Data written to inode at offset %llu. "
1862 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1863 (unsigned long long)pos, *copied,
1864 le16_to_cpu(di->id2.i_data.id_count),
1865 le16_to_cpu(di->i_dyn_features));
1868 int ocfs2_write_end_nolock(struct address_space *mapping,
1869 loff_t pos, unsigned len, unsigned copied,
1870 struct page *page, void *fsdata)
1873 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1874 struct inode *inode = mapping->host;
1875 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1876 struct ocfs2_write_ctxt *wc = fsdata;
1877 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1878 handle_t *handle = wc->w_handle;
1879 struct page *tmppage;
1881 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1882 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1883 goto out_write_size;
1886 if (unlikely(copied < len)) {
1887 if (!PageUptodate(wc->w_target_page))
1890 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1893 flush_dcache_page(wc->w_target_page);
1895 for(i = 0; i < wc->w_num_pages; i++) {
1896 tmppage = wc->w_pages[i];
1898 if (tmppage == wc->w_target_page) {
1899 from = wc->w_target_from;
1900 to = wc->w_target_to;
1902 BUG_ON(from > PAGE_CACHE_SIZE ||
1903 to > PAGE_CACHE_SIZE ||
1907 * Pages adjacent to the target (if any) imply
1908 * a hole-filling write in which case we want
1909 * to flush their entire range.
1912 to = PAGE_CACHE_SIZE;
1915 if (page_has_buffers(tmppage)) {
1916 if (ocfs2_should_order_data(inode))
1917 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1918 block_commit_write(tmppage, from, to);
1924 if (pos > inode->i_size) {
1925 i_size_write(inode, pos);
1926 mark_inode_dirty(inode);
1928 inode->i_blocks = ocfs2_inode_sector_count(inode);
1929 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1930 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1931 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1932 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1933 ocfs2_journal_dirty(handle, wc->w_di_bh);
1935 ocfs2_commit_trans(osb, handle);
1937 ocfs2_run_deallocs(osb, &wc->w_dealloc);
1939 ocfs2_free_write_ctxt(wc);
1944 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
1945 loff_t pos, unsigned len, unsigned copied,
1946 struct page *page, void *fsdata)
1949 struct inode *inode = mapping->host;
1951 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
1953 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1954 ocfs2_inode_unlock(inode, 1);
1959 const struct address_space_operations ocfs2_aops = {
1960 .readpage = ocfs2_readpage,
1961 .readpages = ocfs2_readpages,
1962 .writepage = ocfs2_writepage,
1963 .write_begin = ocfs2_write_begin,
1964 .write_end = ocfs2_write_end,
1966 .sync_page = block_sync_page,
1967 .direct_IO = ocfs2_direct_IO,
1968 .invalidatepage = ocfs2_invalidatepage,
1969 .releasepage = ocfs2_releasepage,
1970 .migratepage = buffer_migrate_page,
1971 .is_partially_uptodate = block_is_partially_uptodate,