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(inode, OCFS2_I(inode)->ip_blkno,
72 &bh, OCFS2_BH_CACHED);
77 fe = (struct ocfs2_dinode *) bh->b_data;
79 if (!OCFS2_IS_VALID_DINODE(fe)) {
80 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
81 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
86 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
87 le32_to_cpu(fe->i_clusters))) {
88 mlog(ML_ERROR, "block offset is outside the allocated size: "
89 "%llu\n", (unsigned long long)iblock);
93 /* We don't use the page cache to create symlink data, so if
94 * need be, copy it over from the buffer cache. */
95 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
96 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
98 buffer_cache_bh = sb_getblk(osb->sb, blkno);
99 if (!buffer_cache_bh) {
100 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
104 /* we haven't locked out transactions, so a commit
105 * could've happened. Since we've got a reference on
106 * the bh, even if it commits while we're doing the
107 * copy, the data is still good. */
108 if (buffer_jbd(buffer_cache_bh)
109 && ocfs2_inode_is_new(inode)) {
110 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
112 mlog(ML_ERROR, "couldn't kmap!\n");
115 memcpy(kaddr + (bh_result->b_size * iblock),
116 buffer_cache_bh->b_data,
118 kunmap_atomic(kaddr, KM_USER0);
119 set_buffer_uptodate(bh_result);
121 brelse(buffer_cache_bh);
124 map_bh(bh_result, inode->i_sb,
125 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
136 static int ocfs2_get_block(struct inode *inode, sector_t iblock,
137 struct buffer_head *bh_result, int create)
140 unsigned int ext_flags;
141 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
142 u64 p_blkno, count, past_eof;
143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
145 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
146 (unsigned long long)iblock, bh_result, create);
148 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
149 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
150 inode, inode->i_ino);
152 if (S_ISLNK(inode->i_mode)) {
153 /* this always does I/O for some reason. */
154 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
158 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
161 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
162 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
163 (unsigned long long)p_blkno);
167 if (max_blocks < count)
171 * ocfs2 never allocates in this function - the only time we
172 * need to use BH_New is when we're extending i_size on a file
173 * system which doesn't support holes, in which case BH_New
174 * allows block_prepare_write() to zero.
176 * If we see this on a sparse file system, then a truncate has
177 * raced us and removed the cluster. In this case, we clear
178 * the buffers dirty and uptodate bits and let the buffer code
179 * ignore it as a hole.
181 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
182 clear_buffer_dirty(bh_result);
183 clear_buffer_uptodate(bh_result);
187 /* Treat the unwritten extent as a hole for zeroing purposes. */
188 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
189 map_bh(bh_result, inode->i_sb, p_blkno);
191 bh_result->b_size = count << inode->i_blkbits;
193 if (!ocfs2_sparse_alloc(osb)) {
197 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
198 (unsigned long long)iblock,
199 (unsigned long long)p_blkno,
200 (unsigned long long)OCFS2_I(inode)->ip_blkno);
201 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
205 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
206 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
207 (unsigned long long)past_eof);
209 if (create && (iblock >= past_eof))
210 set_buffer_new(bh_result);
221 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
222 struct buffer_head *di_bh)
226 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
228 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
229 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
234 size = i_size_read(inode);
236 if (size > PAGE_CACHE_SIZE ||
237 size > ocfs2_max_inline_data(inode->i_sb)) {
238 ocfs2_error(inode->i_sb,
239 "Inode %llu has with inline data has bad size: %Lu",
240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
241 (unsigned long long)size);
245 kaddr = kmap_atomic(page, KM_USER0);
247 memcpy(kaddr, di->id2.i_data.id_data, size);
248 /* Clear the remaining part of the page */
249 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
250 flush_dcache_page(page);
251 kunmap_atomic(kaddr, KM_USER0);
253 SetPageUptodate(page);
258 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
261 struct buffer_head *di_bh = NULL;
263 BUG_ON(!PageLocked(page));
264 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
266 ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &di_bh,
273 ret = ocfs2_read_inline_data(inode, page, di_bh);
281 static int ocfs2_readpage(struct file *file, struct page *page)
283 struct inode *inode = page->mapping->host;
284 struct ocfs2_inode_info *oi = OCFS2_I(inode);
285 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
288 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
290 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
292 if (ret == AOP_TRUNCATED_PAGE)
298 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
299 ret = AOP_TRUNCATED_PAGE;
300 goto out_inode_unlock;
304 * i_size might have just been updated as we grabed the meta lock. We
305 * might now be discovering a truncate that hit on another node.
306 * block_read_full_page->get_block freaks out if it is asked to read
307 * beyond the end of a file, so we check here. Callers
308 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
309 * and notice that the page they just read isn't needed.
311 * XXX sys_readahead() seems to get that wrong?
313 if (start >= i_size_read(inode)) {
314 zero_user(page, 0, PAGE_SIZE);
315 SetPageUptodate(page);
320 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
321 ret = ocfs2_readpage_inline(inode, page);
323 ret = block_read_full_page(page, ocfs2_get_block);
327 up_read(&OCFS2_I(inode)->ip_alloc_sem);
329 ocfs2_inode_unlock(inode, 0);
338 * This is used only for read-ahead. Failures or difficult to handle
339 * situations are safe to ignore.
341 * Right now, we don't bother with BH_Boundary - in-inode extent lists
342 * are quite large (243 extents on 4k blocks), so most inodes don't
343 * grow out to a tree. If need be, detecting boundary extents could
344 * trivially be added in a future version of ocfs2_get_block().
346 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
347 struct list_head *pages, unsigned nr_pages)
350 struct inode *inode = mapping->host;
351 struct ocfs2_inode_info *oi = OCFS2_I(inode);
356 * Use the nonblocking flag for the dlm code to avoid page
357 * lock inversion, but don't bother with retrying.
359 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
363 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
364 ocfs2_inode_unlock(inode, 0);
369 * Don't bother with inline-data. There isn't anything
370 * to read-ahead in that case anyway...
372 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
376 * Check whether a remote node truncated this file - we just
377 * drop out in that case as it's not worth handling here.
379 last = list_entry(pages->prev, struct page, lru);
380 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
381 if (start >= i_size_read(inode))
384 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
387 up_read(&oi->ip_alloc_sem);
388 ocfs2_inode_unlock(inode, 0);
393 /* Note: Because we don't support holes, our allocation has
394 * already happened (allocation writes zeros to the file data)
395 * so we don't have to worry about ordered writes in
398 * ->writepage is called during the process of invalidating the page cache
399 * during blocked lock processing. It can't block on any cluster locks
400 * to during block mapping. It's relying on the fact that the block
401 * mapping can't have disappeared under the dirty pages that it is
402 * being asked to write back.
404 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
408 mlog_entry("(0x%p)\n", page);
410 ret = block_write_full_page(page, ocfs2_get_block, wbc);
418 * This is called from ocfs2_write_zero_page() which has handled it's
419 * own cluster locking and has ensured allocation exists for those
420 * blocks to be written.
422 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
423 unsigned from, unsigned to)
427 ret = block_prepare_write(page, from, to, ocfs2_get_block);
432 /* Taken from ext3. We don't necessarily need the full blown
433 * functionality yet, but IMHO it's better to cut and paste the whole
434 * thing so we can avoid introducing our own bugs (and easily pick up
435 * their fixes when they happen) --Mark */
436 int walk_page_buffers( handle_t *handle,
437 struct buffer_head *head,
441 int (*fn)( handle_t *handle,
442 struct buffer_head *bh))
444 struct buffer_head *bh;
445 unsigned block_start, block_end;
446 unsigned blocksize = head->b_size;
448 struct buffer_head *next;
450 for ( bh = head, block_start = 0;
451 ret == 0 && (bh != head || !block_start);
452 block_start = block_end, bh = next)
454 next = bh->b_this_page;
455 block_end = block_start + blocksize;
456 if (block_end <= from || block_start >= to) {
457 if (partial && !buffer_uptodate(bh))
461 err = (*fn)(handle, bh);
468 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
473 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
477 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
478 if (IS_ERR(handle)) {
484 if (ocfs2_should_order_data(inode)) {
485 ret = ocfs2_jbd2_file_inode(handle, inode);
486 #ifdef CONFIG_OCFS2_COMPAT_JBD
487 ret = walk_page_buffers(handle,
490 ocfs2_journal_dirty_data);
498 ocfs2_commit_trans(osb, handle);
499 handle = ERR_PTR(ret);
504 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
509 struct inode *inode = mapping->host;
511 mlog_entry("(block = %llu)\n", (unsigned long long)block);
513 /* We don't need to lock journal system files, since they aren't
514 * accessed concurrently from multiple nodes.
516 if (!INODE_JOURNAL(inode)) {
517 err = ocfs2_inode_lock(inode, NULL, 0);
523 down_read(&OCFS2_I(inode)->ip_alloc_sem);
526 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
527 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
530 if (!INODE_JOURNAL(inode)) {
531 up_read(&OCFS2_I(inode)->ip_alloc_sem);
532 ocfs2_inode_unlock(inode, 0);
536 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
537 (unsigned long long)block);
543 status = err ? 0 : p_blkno;
545 mlog_exit((int)status);
551 * TODO: Make this into a generic get_blocks function.
553 * From do_direct_io in direct-io.c:
554 * "So what we do is to permit the ->get_blocks function to populate
555 * bh.b_size with the size of IO which is permitted at this offset and
558 * This function is called directly from get_more_blocks in direct-io.c.
560 * called like this: dio->get_blocks(dio->inode, fs_startblk,
561 * fs_count, map_bh, dio->rw == WRITE);
563 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
564 struct buffer_head *bh_result, int create)
567 u64 p_blkno, inode_blocks, contig_blocks;
568 unsigned int ext_flags;
569 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
570 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
572 /* This function won't even be called if the request isn't all
573 * nicely aligned and of the right size, so there's no need
574 * for us to check any of that. */
576 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
579 * Any write past EOF is not allowed because we'd be extending.
581 if (create && (iblock + max_blocks) > inode_blocks) {
586 /* This figures out the size of the next contiguous block, and
587 * our logical offset */
588 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
589 &contig_blocks, &ext_flags);
591 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
592 (unsigned long long)iblock);
597 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno && create) {
598 ocfs2_error(inode->i_sb,
599 "Inode %llu has a hole at block %llu\n",
600 (unsigned long long)OCFS2_I(inode)->ip_blkno,
601 (unsigned long long)iblock);
607 * get_more_blocks() expects us to describe a hole by clearing
608 * the mapped bit on bh_result().
610 * Consider an unwritten extent as a hole.
612 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
613 map_bh(bh_result, inode->i_sb, p_blkno);
616 * ocfs2_prepare_inode_for_write() should have caught
617 * the case where we'd be filling a hole and triggered
618 * a buffered write instead.
626 clear_buffer_mapped(bh_result);
629 /* make sure we don't map more than max_blocks blocks here as
630 that's all the kernel will handle at this point. */
631 if (max_blocks < contig_blocks)
632 contig_blocks = max_blocks;
633 bh_result->b_size = contig_blocks << blocksize_bits;
639 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
640 * particularly interested in the aio/dio case. Like the core uses
641 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
642 * truncation on another.
644 static void ocfs2_dio_end_io(struct kiocb *iocb,
649 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
652 /* this io's submitter should not have unlocked this before we could */
653 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
655 ocfs2_iocb_clear_rw_locked(iocb);
657 level = ocfs2_iocb_rw_locked_level(iocb);
659 up_read(&inode->i_alloc_sem);
660 ocfs2_rw_unlock(inode, level);
664 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
665 * from ext3. PageChecked() bits have been removed as OCFS2 does not
666 * do journalled data.
668 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
670 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
672 jbd2_journal_invalidatepage(journal, page, offset);
675 static int ocfs2_releasepage(struct page *page, gfp_t wait)
677 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
679 if (!page_has_buffers(page))
681 return jbd2_journal_try_to_free_buffers(journal, page, wait);
684 static ssize_t ocfs2_direct_IO(int rw,
686 const struct iovec *iov,
688 unsigned long nr_segs)
690 struct file *file = iocb->ki_filp;
691 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
697 * Fallback to buffered I/O if we see an inode without
700 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
703 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
704 inode->i_sb->s_bdev, iov, offset,
706 ocfs2_direct_IO_get_blocks,
713 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
718 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
720 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
723 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
725 cluster_start = cpos % cpp;
726 cluster_start = cluster_start << osb->s_clustersize_bits;
728 cluster_end = cluster_start + osb->s_clustersize;
731 BUG_ON(cluster_start > PAGE_SIZE);
732 BUG_ON(cluster_end > PAGE_SIZE);
735 *start = cluster_start;
741 * 'from' and 'to' are the region in the page to avoid zeroing.
743 * If pagesize > clustersize, this function will avoid zeroing outside
744 * of the cluster boundary.
746 * from == to == 0 is code for "zero the entire cluster region"
748 static void ocfs2_clear_page_regions(struct page *page,
749 struct ocfs2_super *osb, u32 cpos,
750 unsigned from, unsigned to)
753 unsigned int cluster_start, cluster_end;
755 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
757 kaddr = kmap_atomic(page, KM_USER0);
760 if (from > cluster_start)
761 memset(kaddr + cluster_start, 0, from - cluster_start);
762 if (to < cluster_end)
763 memset(kaddr + to, 0, cluster_end - to);
765 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
768 kunmap_atomic(kaddr, KM_USER0);
772 * Nonsparse file systems fully allocate before we get to the write
773 * code. This prevents ocfs2_write() from tagging the write as an
774 * allocating one, which means ocfs2_map_page_blocks() might try to
775 * read-in the blocks at the tail of our file. Avoid reading them by
776 * testing i_size against each block offset.
778 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
779 unsigned int block_start)
781 u64 offset = page_offset(page) + block_start;
783 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
786 if (i_size_read(inode) > offset)
793 * Some of this taken from block_prepare_write(). We already have our
794 * mapping by now though, and the entire write will be allocating or
795 * it won't, so not much need to use BH_New.
797 * This will also skip zeroing, which is handled externally.
799 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
800 struct inode *inode, unsigned int from,
801 unsigned int to, int new)
804 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
805 unsigned int block_end, block_start;
806 unsigned int bsize = 1 << inode->i_blkbits;
808 if (!page_has_buffers(page))
809 create_empty_buffers(page, bsize, 0);
811 head = page_buffers(page);
812 for (bh = head, block_start = 0; bh != head || !block_start;
813 bh = bh->b_this_page, block_start += bsize) {
814 block_end = block_start + bsize;
816 clear_buffer_new(bh);
819 * Ignore blocks outside of our i/o range -
820 * they may belong to unallocated clusters.
822 if (block_start >= to || block_end <= from) {
823 if (PageUptodate(page))
824 set_buffer_uptodate(bh);
829 * For an allocating write with cluster size >= page
830 * size, we always write the entire page.
835 if (!buffer_mapped(bh)) {
836 map_bh(bh, inode->i_sb, *p_blkno);
837 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
840 if (PageUptodate(page)) {
841 if (!buffer_uptodate(bh))
842 set_buffer_uptodate(bh);
843 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
845 ocfs2_should_read_blk(inode, page, block_start) &&
846 (block_start < from || block_end > to)) {
847 ll_rw_block(READ, 1, &bh);
851 *p_blkno = *p_blkno + 1;
855 * If we issued read requests - let them complete.
857 while(wait_bh > wait) {
858 wait_on_buffer(*--wait_bh);
859 if (!buffer_uptodate(*wait_bh))
863 if (ret == 0 || !new)
867 * If we get -EIO above, zero out any newly allocated blocks
868 * to avoid exposing stale data.
873 block_end = block_start + bsize;
874 if (block_end <= from)
876 if (block_start >= to)
879 zero_user(page, block_start, bh->b_size);
880 set_buffer_uptodate(bh);
881 mark_buffer_dirty(bh);
884 block_start = block_end;
885 bh = bh->b_this_page;
886 } while (bh != head);
891 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
892 #define OCFS2_MAX_CTXT_PAGES 1
894 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
897 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
900 * Describe the state of a single cluster to be written to.
902 struct ocfs2_write_cluster_desc {
906 * Give this a unique field because c_phys eventually gets
910 unsigned c_unwritten;
913 static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
915 return d->c_new || d->c_unwritten;
918 struct ocfs2_write_ctxt {
919 /* Logical cluster position / len of write */
923 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
926 * This is true if page_size > cluster_size.
928 * It triggers a set of special cases during write which might
929 * have to deal with allocating writes to partial pages.
931 unsigned int w_large_pages;
934 * Pages involved in this write.
936 * w_target_page is the page being written to by the user.
938 * w_pages is an array of pages which always contains
939 * w_target_page, and in the case of an allocating write with
940 * page_size < cluster size, it will contain zero'd and mapped
941 * pages adjacent to w_target_page which need to be written
942 * out in so that future reads from that region will get
945 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
946 unsigned int w_num_pages;
947 struct page *w_target_page;
950 * ocfs2_write_end() uses this to know what the real range to
951 * write in the target should be.
953 unsigned int w_target_from;
954 unsigned int w_target_to;
957 * We could use journal_current_handle() but this is cleaner,
962 struct buffer_head *w_di_bh;
964 struct ocfs2_cached_dealloc_ctxt w_dealloc;
967 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
971 for(i = 0; i < num_pages; i++) {
973 unlock_page(pages[i]);
974 mark_page_accessed(pages[i]);
975 page_cache_release(pages[i]);
980 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
982 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
988 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
989 struct ocfs2_super *osb, loff_t pos,
990 unsigned len, struct buffer_head *di_bh)
993 struct ocfs2_write_ctxt *wc;
995 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
999 wc->w_cpos = pos >> osb->s_clustersize_bits;
1000 cend = (pos + len - 1) >> osb->s_clustersize_bits;
1001 wc->w_clen = cend - wc->w_cpos + 1;
1003 wc->w_di_bh = di_bh;
1005 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
1006 wc->w_large_pages = 1;
1008 wc->w_large_pages = 0;
1010 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1018 * If a page has any new buffers, zero them out here, and mark them uptodate
1019 * and dirty so they'll be written out (in order to prevent uninitialised
1020 * block data from leaking). And clear the new bit.
1022 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1024 unsigned int block_start, block_end;
1025 struct buffer_head *head, *bh;
1027 BUG_ON(!PageLocked(page));
1028 if (!page_has_buffers(page))
1031 bh = head = page_buffers(page);
1034 block_end = block_start + bh->b_size;
1036 if (buffer_new(bh)) {
1037 if (block_end > from && block_start < to) {
1038 if (!PageUptodate(page)) {
1039 unsigned start, end;
1041 start = max(from, block_start);
1042 end = min(to, block_end);
1044 zero_user_segment(page, start, end);
1045 set_buffer_uptodate(bh);
1048 clear_buffer_new(bh);
1049 mark_buffer_dirty(bh);
1053 block_start = block_end;
1054 bh = bh->b_this_page;
1055 } while (bh != head);
1059 * Only called when we have a failure during allocating write to write
1060 * zero's to the newly allocated region.
1062 static void ocfs2_write_failure(struct inode *inode,
1063 struct ocfs2_write_ctxt *wc,
1064 loff_t user_pos, unsigned user_len)
1067 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1068 to = user_pos + user_len;
1069 struct page *tmppage;
1071 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1073 for(i = 0; i < wc->w_num_pages; i++) {
1074 tmppage = wc->w_pages[i];
1076 if (page_has_buffers(tmppage)) {
1077 if (ocfs2_should_order_data(inode)) {
1078 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1079 #ifdef CONFIG_OCFS2_COMPAT_JBD
1080 walk_page_buffers(wc->w_handle,
1081 page_buffers(tmppage),
1083 ocfs2_journal_dirty_data);
1087 block_commit_write(tmppage, from, to);
1092 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1093 struct ocfs2_write_ctxt *wc,
1094 struct page *page, u32 cpos,
1095 loff_t user_pos, unsigned user_len,
1099 unsigned int map_from = 0, map_to = 0;
1100 unsigned int cluster_start, cluster_end;
1101 unsigned int user_data_from = 0, user_data_to = 0;
1103 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1104 &cluster_start, &cluster_end);
1106 if (page == wc->w_target_page) {
1107 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1108 map_to = map_from + user_len;
1111 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1112 cluster_start, cluster_end,
1115 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1116 map_from, map_to, new);
1122 user_data_from = map_from;
1123 user_data_to = map_to;
1125 map_from = cluster_start;
1126 map_to = cluster_end;
1130 * If we haven't allocated the new page yet, we
1131 * shouldn't be writing it out without copying user
1132 * data. This is likely a math error from the caller.
1136 map_from = cluster_start;
1137 map_to = cluster_end;
1139 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1140 cluster_start, cluster_end, new);
1148 * Parts of newly allocated pages need to be zero'd.
1150 * Above, we have also rewritten 'to' and 'from' - as far as
1151 * the rest of the function is concerned, the entire cluster
1152 * range inside of a page needs to be written.
1154 * We can skip this if the page is up to date - it's already
1155 * been zero'd from being read in as a hole.
1157 if (new && !PageUptodate(page))
1158 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1159 cpos, user_data_from, user_data_to);
1161 flush_dcache_page(page);
1168 * This function will only grab one clusters worth of pages.
1170 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1171 struct ocfs2_write_ctxt *wc,
1172 u32 cpos, loff_t user_pos, int new,
1173 struct page *mmap_page)
1176 unsigned long start, target_index, index;
1177 struct inode *inode = mapping->host;
1179 target_index = user_pos >> PAGE_CACHE_SHIFT;
1182 * Figure out how many pages we'll be manipulating here. For
1183 * non allocating write, we just change the one
1184 * page. Otherwise, we'll need a whole clusters worth.
1187 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1188 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1190 wc->w_num_pages = 1;
1191 start = target_index;
1194 for(i = 0; i < wc->w_num_pages; i++) {
1197 if (index == target_index && mmap_page) {
1199 * ocfs2_pagemkwrite() is a little different
1200 * and wants us to directly use the page
1203 lock_page(mmap_page);
1205 if (mmap_page->mapping != mapping) {
1206 unlock_page(mmap_page);
1208 * Sanity check - the locking in
1209 * ocfs2_pagemkwrite() should ensure
1210 * that this code doesn't trigger.
1217 page_cache_get(mmap_page);
1218 wc->w_pages[i] = mmap_page;
1220 wc->w_pages[i] = find_or_create_page(mapping, index,
1222 if (!wc->w_pages[i]) {
1229 if (index == target_index)
1230 wc->w_target_page = wc->w_pages[i];
1237 * Prepare a single cluster for write one cluster into the file.
1239 static int ocfs2_write_cluster(struct address_space *mapping,
1240 u32 phys, unsigned int unwritten,
1241 struct ocfs2_alloc_context *data_ac,
1242 struct ocfs2_alloc_context *meta_ac,
1243 struct ocfs2_write_ctxt *wc, u32 cpos,
1244 loff_t user_pos, unsigned user_len)
1246 int ret, i, new, should_zero = 0;
1247 u64 v_blkno, p_blkno;
1248 struct inode *inode = mapping->host;
1249 struct ocfs2_extent_tree et;
1251 new = phys == 0 ? 1 : 0;
1252 if (new || unwritten)
1259 * This is safe to call with the page locks - it won't take
1260 * any additional semaphores or cluster locks.
1263 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1264 &tmp_pos, 1, 0, wc->w_di_bh,
1265 wc->w_handle, data_ac,
1268 * This shouldn't happen because we must have already
1269 * calculated the correct meta data allocation required. The
1270 * internal tree allocation code should know how to increase
1271 * transaction credits itself.
1273 * If need be, we could handle -EAGAIN for a
1274 * RESTART_TRANS here.
1276 mlog_bug_on_msg(ret == -EAGAIN,
1277 "Inode %llu: EAGAIN return during allocation.\n",
1278 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1283 } else if (unwritten) {
1284 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1285 ret = ocfs2_mark_extent_written(inode, &et,
1286 wc->w_handle, cpos, 1, phys,
1287 meta_ac, &wc->w_dealloc);
1295 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1297 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1300 * The only reason this should fail is due to an inability to
1301 * find the extent added.
1303 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1306 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1307 "at logical block %llu",
1308 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1309 (unsigned long long)v_blkno);
1313 BUG_ON(p_blkno == 0);
1315 for(i = 0; i < wc->w_num_pages; i++) {
1318 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1319 wc->w_pages[i], cpos,
1330 * We only have cleanup to do in case of allocating write.
1333 ocfs2_write_failure(inode, wc, user_pos, user_len);
1340 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1341 struct ocfs2_alloc_context *data_ac,
1342 struct ocfs2_alloc_context *meta_ac,
1343 struct ocfs2_write_ctxt *wc,
1344 loff_t pos, unsigned len)
1348 unsigned int local_len = len;
1349 struct ocfs2_write_cluster_desc *desc;
1350 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1352 for (i = 0; i < wc->w_clen; i++) {
1353 desc = &wc->w_desc[i];
1356 * We have to make sure that the total write passed in
1357 * doesn't extend past a single cluster.
1360 cluster_off = pos & (osb->s_clustersize - 1);
1361 if ((cluster_off + local_len) > osb->s_clustersize)
1362 local_len = osb->s_clustersize - cluster_off;
1364 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1365 desc->c_unwritten, data_ac, meta_ac,
1366 wc, desc->c_cpos, pos, local_len);
1382 * ocfs2_write_end() wants to know which parts of the target page it
1383 * should complete the write on. It's easiest to compute them ahead of
1384 * time when a more complete view of the write is available.
1386 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1387 struct ocfs2_write_ctxt *wc,
1388 loff_t pos, unsigned len, int alloc)
1390 struct ocfs2_write_cluster_desc *desc;
1392 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1393 wc->w_target_to = wc->w_target_from + len;
1399 * Allocating write - we may have different boundaries based
1400 * on page size and cluster size.
1402 * NOTE: We can no longer compute one value from the other as
1403 * the actual write length and user provided length may be
1407 if (wc->w_large_pages) {
1409 * We only care about the 1st and last cluster within
1410 * our range and whether they should be zero'd or not. Either
1411 * value may be extended out to the start/end of a
1412 * newly allocated cluster.
1414 desc = &wc->w_desc[0];
1415 if (ocfs2_should_zero_cluster(desc))
1416 ocfs2_figure_cluster_boundaries(osb,
1421 desc = &wc->w_desc[wc->w_clen - 1];
1422 if (ocfs2_should_zero_cluster(desc))
1423 ocfs2_figure_cluster_boundaries(osb,
1428 wc->w_target_from = 0;
1429 wc->w_target_to = PAGE_CACHE_SIZE;
1434 * Populate each single-cluster write descriptor in the write context
1435 * with information about the i/o to be done.
1437 * Returns the number of clusters that will have to be allocated, as
1438 * well as a worst case estimate of the number of extent records that
1439 * would have to be created during a write to an unwritten region.
1441 static int ocfs2_populate_write_desc(struct inode *inode,
1442 struct ocfs2_write_ctxt *wc,
1443 unsigned int *clusters_to_alloc,
1444 unsigned int *extents_to_split)
1447 struct ocfs2_write_cluster_desc *desc;
1448 unsigned int num_clusters = 0;
1449 unsigned int ext_flags = 0;
1453 *clusters_to_alloc = 0;
1454 *extents_to_split = 0;
1456 for (i = 0; i < wc->w_clen; i++) {
1457 desc = &wc->w_desc[i];
1458 desc->c_cpos = wc->w_cpos + i;
1460 if (num_clusters == 0) {
1462 * Need to look up the next extent record.
1464 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1465 &num_clusters, &ext_flags);
1472 * Assume worst case - that we're writing in
1473 * the middle of the extent.
1475 * We can assume that the write proceeds from
1476 * left to right, in which case the extent
1477 * insert code is smart enough to coalesce the
1478 * next splits into the previous records created.
1480 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1481 *extents_to_split = *extents_to_split + 2;
1484 * Only increment phys if it doesn't describe
1490 desc->c_phys = phys;
1493 *clusters_to_alloc = *clusters_to_alloc + 1;
1495 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1496 desc->c_unwritten = 1;
1506 static int ocfs2_write_begin_inline(struct address_space *mapping,
1507 struct inode *inode,
1508 struct ocfs2_write_ctxt *wc)
1511 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1514 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1516 page = find_or_create_page(mapping, 0, GFP_NOFS);
1523 * If we don't set w_num_pages then this page won't get unlocked
1524 * and freed on cleanup of the write context.
1526 wc->w_pages[0] = wc->w_target_page = page;
1527 wc->w_num_pages = 1;
1529 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1530 if (IS_ERR(handle)) {
1531 ret = PTR_ERR(handle);
1536 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1537 OCFS2_JOURNAL_ACCESS_WRITE);
1539 ocfs2_commit_trans(osb, handle);
1545 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1546 ocfs2_set_inode_data_inline(inode, di);
1548 if (!PageUptodate(page)) {
1549 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1551 ocfs2_commit_trans(osb, handle);
1557 wc->w_handle = handle;
1562 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1564 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1566 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1571 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1572 struct inode *inode, loff_t pos,
1573 unsigned len, struct page *mmap_page,
1574 struct ocfs2_write_ctxt *wc)
1576 int ret, written = 0;
1577 loff_t end = pos + len;
1578 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1580 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1581 (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos,
1582 oi->ip_dyn_features);
1585 * Handle inodes which already have inline data 1st.
1587 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1588 if (mmap_page == NULL &&
1589 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1590 goto do_inline_write;
1593 * The write won't fit - we have to give this inode an
1594 * inline extent list now.
1596 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1603 * Check whether the inode can accept inline data.
1605 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1609 * Check whether the write can fit.
1611 if (mmap_page || end > ocfs2_max_inline_data(inode->i_sb))
1615 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1622 * This signals to the caller that the data can be written
1627 return written ? written : ret;
1631 * This function only does anything for file systems which can't
1632 * handle sparse files.
1634 * What we want to do here is fill in any hole between the current end
1635 * of allocation and the end of our write. That way the rest of the
1636 * write path can treat it as an non-allocating write, which has no
1637 * special case code for sparse/nonsparse files.
1639 static int ocfs2_expand_nonsparse_inode(struct inode *inode, loff_t pos,
1641 struct ocfs2_write_ctxt *wc)
1644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645 loff_t newsize = pos + len;
1647 if (ocfs2_sparse_alloc(osb))
1650 if (newsize <= i_size_read(inode))
1653 ret = ocfs2_extend_no_holes(inode, newsize, newsize - len);
1660 int ocfs2_write_begin_nolock(struct address_space *mapping,
1661 loff_t pos, unsigned len, unsigned flags,
1662 struct page **pagep, void **fsdata,
1663 struct buffer_head *di_bh, struct page *mmap_page)
1665 int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
1666 unsigned int clusters_to_alloc, extents_to_split;
1667 struct ocfs2_write_ctxt *wc;
1668 struct inode *inode = mapping->host;
1669 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1670 struct ocfs2_dinode *di;
1671 struct ocfs2_alloc_context *data_ac = NULL;
1672 struct ocfs2_alloc_context *meta_ac = NULL;
1674 struct ocfs2_extent_tree et;
1676 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1682 if (ocfs2_supports_inline_data(osb)) {
1683 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1695 ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc);
1701 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1708 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1711 * We set w_target_from, w_target_to here so that
1712 * ocfs2_write_end() knows which range in the target page to
1713 * write out. An allocation requires that we write the entire
1716 if (clusters_to_alloc || extents_to_split) {
1718 * XXX: We are stretching the limits of
1719 * ocfs2_lock_allocators(). It greatly over-estimates
1720 * the work to be done.
1722 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1723 " clusters_to_add = %u, extents_to_split = %u\n",
1724 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1725 (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
1726 clusters_to_alloc, extents_to_split);
1728 ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
1729 ret = ocfs2_lock_allocators(inode, &et,
1730 clusters_to_alloc, extents_to_split,
1731 &data_ac, &meta_ac);
1737 credits = ocfs2_calc_extend_credits(inode->i_sb,
1743 ocfs2_set_target_boundaries(osb, wc, pos, len,
1744 clusters_to_alloc + extents_to_split);
1746 handle = ocfs2_start_trans(osb, credits);
1747 if (IS_ERR(handle)) {
1748 ret = PTR_ERR(handle);
1753 wc->w_handle = handle;
1756 * We don't want this to fail in ocfs2_write_end(), so do it
1759 ret = ocfs2_journal_access(handle, inode, wc->w_di_bh,
1760 OCFS2_JOURNAL_ACCESS_WRITE);
1767 * Fill our page array first. That way we've grabbed enough so
1768 * that we can zero and flush if we error after adding the
1771 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
1772 clusters_to_alloc + extents_to_split,
1779 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1787 ocfs2_free_alloc_context(data_ac);
1789 ocfs2_free_alloc_context(meta_ac);
1792 *pagep = wc->w_target_page;
1796 ocfs2_commit_trans(osb, handle);
1799 ocfs2_free_write_ctxt(wc);
1802 ocfs2_free_alloc_context(data_ac);
1804 ocfs2_free_alloc_context(meta_ac);
1808 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1809 loff_t pos, unsigned len, unsigned flags,
1810 struct page **pagep, void **fsdata)
1813 struct buffer_head *di_bh = NULL;
1814 struct inode *inode = mapping->host;
1816 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1823 * Take alloc sem here to prevent concurrent lookups. That way
1824 * the mapping, zeroing and tree manipulation within
1825 * ocfs2_write() will be safe against ->readpage(). This
1826 * should also serve to lock out allocation from a shared
1829 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1831 ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
1832 fsdata, di_bh, NULL);
1843 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1846 ocfs2_inode_unlock(inode, 1);
1851 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1852 unsigned len, unsigned *copied,
1853 struct ocfs2_dinode *di,
1854 struct ocfs2_write_ctxt *wc)
1858 if (unlikely(*copied < len)) {
1859 if (!PageUptodate(wc->w_target_page)) {
1865 kaddr = kmap_atomic(wc->w_target_page, KM_USER0);
1866 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1867 kunmap_atomic(kaddr, KM_USER0);
1869 mlog(0, "Data written to inode at offset %llu. "
1870 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1871 (unsigned long long)pos, *copied,
1872 le16_to_cpu(di->id2.i_data.id_count),
1873 le16_to_cpu(di->i_dyn_features));
1876 int ocfs2_write_end_nolock(struct address_space *mapping,
1877 loff_t pos, unsigned len, unsigned copied,
1878 struct page *page, void *fsdata)
1881 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1882 struct inode *inode = mapping->host;
1883 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1884 struct ocfs2_write_ctxt *wc = fsdata;
1885 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1886 handle_t *handle = wc->w_handle;
1887 struct page *tmppage;
1889 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1890 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1891 goto out_write_size;
1894 if (unlikely(copied < len)) {
1895 if (!PageUptodate(wc->w_target_page))
1898 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1901 flush_dcache_page(wc->w_target_page);
1903 for(i = 0; i < wc->w_num_pages; i++) {
1904 tmppage = wc->w_pages[i];
1906 if (tmppage == wc->w_target_page) {
1907 from = wc->w_target_from;
1908 to = wc->w_target_to;
1910 BUG_ON(from > PAGE_CACHE_SIZE ||
1911 to > PAGE_CACHE_SIZE ||
1915 * Pages adjacent to the target (if any) imply
1916 * a hole-filling write in which case we want
1917 * to flush their entire range.
1920 to = PAGE_CACHE_SIZE;
1923 if (page_has_buffers(tmppage)) {
1924 if (ocfs2_should_order_data(inode)) {
1925 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1926 #ifdef CONFIG_OCFS2_COMPAT_JBD
1927 walk_page_buffers(wc->w_handle,
1928 page_buffers(tmppage),
1930 ocfs2_journal_dirty_data);
1933 block_commit_write(tmppage, from, to);
1939 if (pos > inode->i_size) {
1940 i_size_write(inode, pos);
1941 mark_inode_dirty(inode);
1943 inode->i_blocks = ocfs2_inode_sector_count(inode);
1944 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1945 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1946 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1947 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1948 ocfs2_journal_dirty(handle, wc->w_di_bh);
1950 ocfs2_commit_trans(osb, handle);
1952 ocfs2_run_deallocs(osb, &wc->w_dealloc);
1954 ocfs2_free_write_ctxt(wc);
1959 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
1960 loff_t pos, unsigned len, unsigned copied,
1961 struct page *page, void *fsdata)
1964 struct inode *inode = mapping->host;
1966 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
1968 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1969 ocfs2_inode_unlock(inode, 1);
1974 const struct address_space_operations ocfs2_aops = {
1975 .readpage = ocfs2_readpage,
1976 .readpages = ocfs2_readpages,
1977 .writepage = ocfs2_writepage,
1978 .write_begin = ocfs2_write_begin,
1979 .write_end = ocfs2_write_end,
1981 .sync_page = block_sync_page,
1982 .direct_IO = ocfs2_direct_IO,
1983 .invalidatepage = ocfs2_invalidatepage,
1984 .releasepage = ocfs2_releasepage,
1985 .migratepage = buffer_migrate_page,