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"
47 #include "refcounttree.h"
49 #include "buffer_head_io.h"
51 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
52 struct buffer_head *bh_result, int create)
56 struct ocfs2_dinode *fe = NULL;
57 struct buffer_head *bh = NULL;
58 struct buffer_head *buffer_cache_bh = NULL;
59 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
62 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
63 (unsigned long long)iblock, bh_result, create);
65 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
67 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
68 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
69 (unsigned long long)iblock);
73 status = ocfs2_read_inode_block(inode, &bh);
78 fe = (struct ocfs2_dinode *) bh->b_data;
80 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
81 le32_to_cpu(fe->i_clusters))) {
82 mlog(ML_ERROR, "block offset is outside the allocated size: "
83 "%llu\n", (unsigned long long)iblock);
87 /* We don't use the page cache to create symlink data, so if
88 * need be, copy it over from the buffer cache. */
89 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
90 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
92 buffer_cache_bh = sb_getblk(osb->sb, blkno);
93 if (!buffer_cache_bh) {
94 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
98 /* we haven't locked out transactions, so a commit
99 * could've happened. Since we've got a reference on
100 * the bh, even if it commits while we're doing the
101 * copy, the data is still good. */
102 if (buffer_jbd(buffer_cache_bh)
103 && ocfs2_inode_is_new(inode)) {
104 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
106 mlog(ML_ERROR, "couldn't kmap!\n");
109 memcpy(kaddr + (bh_result->b_size * iblock),
110 buffer_cache_bh->b_data,
112 kunmap_atomic(kaddr, KM_USER0);
113 set_buffer_uptodate(bh_result);
115 brelse(buffer_cache_bh);
118 map_bh(bh_result, inode->i_sb,
119 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
130 int ocfs2_get_block(struct inode *inode, sector_t iblock,
131 struct buffer_head *bh_result, int create)
134 unsigned int ext_flags;
135 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
136 u64 p_blkno, count, past_eof;
137 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
139 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
140 (unsigned long long)iblock, bh_result, create);
142 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
143 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
144 inode, inode->i_ino);
146 if (S_ISLNK(inode->i_mode)) {
147 /* this always does I/O for some reason. */
148 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
152 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
155 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
156 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
157 (unsigned long long)p_blkno);
161 if (max_blocks < count)
165 * ocfs2 never allocates in this function - the only time we
166 * need to use BH_New is when we're extending i_size on a file
167 * system which doesn't support holes, in which case BH_New
168 * allows block_prepare_write() to zero.
170 * If we see this on a sparse file system, then a truncate has
171 * raced us and removed the cluster. In this case, we clear
172 * the buffers dirty and uptodate bits and let the buffer code
173 * ignore it as a hole.
175 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
176 clear_buffer_dirty(bh_result);
177 clear_buffer_uptodate(bh_result);
181 /* Treat the unwritten extent as a hole for zeroing purposes. */
182 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
183 map_bh(bh_result, inode->i_sb, p_blkno);
185 bh_result->b_size = count << inode->i_blkbits;
187 if (!ocfs2_sparse_alloc(osb)) {
191 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
192 (unsigned long long)iblock,
193 (unsigned long long)p_blkno,
194 (unsigned long long)OCFS2_I(inode)->ip_blkno);
195 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
200 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
201 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
202 (unsigned long long)past_eof);
204 if (create && (iblock >= past_eof))
205 set_buffer_new(bh_result);
216 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
217 struct buffer_head *di_bh)
221 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
223 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
224 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
225 (unsigned long long)OCFS2_I(inode)->ip_blkno);
229 size = i_size_read(inode);
231 if (size > PAGE_CACHE_SIZE ||
232 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
233 ocfs2_error(inode->i_sb,
234 "Inode %llu has with inline data has bad size: %Lu",
235 (unsigned long long)OCFS2_I(inode)->ip_blkno,
236 (unsigned long long)size);
240 kaddr = kmap_atomic(page, KM_USER0);
242 memcpy(kaddr, di->id2.i_data.id_data, size);
243 /* Clear the remaining part of the page */
244 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
245 flush_dcache_page(page);
246 kunmap_atomic(kaddr, KM_USER0);
248 SetPageUptodate(page);
253 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
256 struct buffer_head *di_bh = NULL;
258 BUG_ON(!PageLocked(page));
259 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
261 ret = ocfs2_read_inode_block(inode, &di_bh);
267 ret = ocfs2_read_inline_data(inode, page, di_bh);
275 static int ocfs2_readpage(struct file *file, struct page *page)
277 struct inode *inode = page->mapping->host;
278 struct ocfs2_inode_info *oi = OCFS2_I(inode);
279 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
282 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
284 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
286 if (ret == AOP_TRUNCATED_PAGE)
292 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
293 ret = AOP_TRUNCATED_PAGE;
294 goto out_inode_unlock;
298 * i_size might have just been updated as we grabed the meta lock. We
299 * might now be discovering a truncate that hit on another node.
300 * block_read_full_page->get_block freaks out if it is asked to read
301 * beyond the end of a file, so we check here. Callers
302 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
303 * and notice that the page they just read isn't needed.
305 * XXX sys_readahead() seems to get that wrong?
307 if (start >= i_size_read(inode)) {
308 zero_user(page, 0, PAGE_SIZE);
309 SetPageUptodate(page);
314 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
315 ret = ocfs2_readpage_inline(inode, page);
317 ret = block_read_full_page(page, ocfs2_get_block);
321 up_read(&OCFS2_I(inode)->ip_alloc_sem);
323 ocfs2_inode_unlock(inode, 0);
332 * This is used only for read-ahead. Failures or difficult to handle
333 * situations are safe to ignore.
335 * Right now, we don't bother with BH_Boundary - in-inode extent lists
336 * are quite large (243 extents on 4k blocks), so most inodes don't
337 * grow out to a tree. If need be, detecting boundary extents could
338 * trivially be added in a future version of ocfs2_get_block().
340 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
341 struct list_head *pages, unsigned nr_pages)
344 struct inode *inode = mapping->host;
345 struct ocfs2_inode_info *oi = OCFS2_I(inode);
350 * Use the nonblocking flag for the dlm code to avoid page
351 * lock inversion, but don't bother with retrying.
353 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
357 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
358 ocfs2_inode_unlock(inode, 0);
363 * Don't bother with inline-data. There isn't anything
364 * to read-ahead in that case anyway...
366 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
370 * Check whether a remote node truncated this file - we just
371 * drop out in that case as it's not worth handling here.
373 last = list_entry(pages->prev, struct page, lru);
374 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
375 if (start >= i_size_read(inode))
378 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
381 up_read(&oi->ip_alloc_sem);
382 ocfs2_inode_unlock(inode, 0);
387 /* Note: Because we don't support holes, our allocation has
388 * already happened (allocation writes zeros to the file data)
389 * so we don't have to worry about ordered writes in
392 * ->writepage is called during the process of invalidating the page cache
393 * during blocked lock processing. It can't block on any cluster locks
394 * to during block mapping. It's relying on the fact that the block
395 * mapping can't have disappeared under the dirty pages that it is
396 * being asked to write back.
398 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
402 mlog_entry("(0x%p)\n", page);
404 ret = block_write_full_page(page, ocfs2_get_block, wbc);
412 * This is called from ocfs2_write_zero_page() which has handled it's
413 * own cluster locking and has ensured allocation exists for those
414 * blocks to be written.
416 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
417 unsigned from, unsigned to)
421 ret = block_prepare_write(page, from, to, ocfs2_get_block);
426 /* Taken from ext3. We don't necessarily need the full blown
427 * functionality yet, but IMHO it's better to cut and paste the whole
428 * thing so we can avoid introducing our own bugs (and easily pick up
429 * their fixes when they happen) --Mark */
430 int walk_page_buffers( handle_t *handle,
431 struct buffer_head *head,
435 int (*fn)( handle_t *handle,
436 struct buffer_head *bh))
438 struct buffer_head *bh;
439 unsigned block_start, block_end;
440 unsigned blocksize = head->b_size;
442 struct buffer_head *next;
444 for ( bh = head, block_start = 0;
445 ret == 0 && (bh != head || !block_start);
446 block_start = block_end, bh = next)
448 next = bh->b_this_page;
449 block_end = block_start + blocksize;
450 if (block_end <= from || block_start >= to) {
451 if (partial && !buffer_uptodate(bh))
455 err = (*fn)(handle, bh);
462 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
467 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
471 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
472 if (IS_ERR(handle)) {
478 if (ocfs2_should_order_data(inode)) {
479 ret = ocfs2_jbd2_file_inode(handle, inode);
486 ocfs2_commit_trans(osb, handle);
487 handle = ERR_PTR(ret);
492 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
497 struct inode *inode = mapping->host;
499 mlog_entry("(block = %llu)\n", (unsigned long long)block);
501 /* We don't need to lock journal system files, since they aren't
502 * accessed concurrently from multiple nodes.
504 if (!INODE_JOURNAL(inode)) {
505 err = ocfs2_inode_lock(inode, NULL, 0);
511 down_read(&OCFS2_I(inode)->ip_alloc_sem);
514 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
515 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
518 if (!INODE_JOURNAL(inode)) {
519 up_read(&OCFS2_I(inode)->ip_alloc_sem);
520 ocfs2_inode_unlock(inode, 0);
524 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
525 (unsigned long long)block);
531 status = err ? 0 : p_blkno;
533 mlog_exit((int)status);
539 * TODO: Make this into a generic get_blocks function.
541 * From do_direct_io in direct-io.c:
542 * "So what we do is to permit the ->get_blocks function to populate
543 * bh.b_size with the size of IO which is permitted at this offset and
546 * This function is called directly from get_more_blocks in direct-io.c.
548 * called like this: dio->get_blocks(dio->inode, fs_startblk,
549 * fs_count, map_bh, dio->rw == WRITE);
551 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
552 struct buffer_head *bh_result, int create)
555 u64 p_blkno, inode_blocks, contig_blocks;
556 unsigned int ext_flags;
557 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
558 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
560 /* This function won't even be called if the request isn't all
561 * nicely aligned and of the right size, so there's no need
562 * for us to check any of that. */
564 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
567 * Any write past EOF is not allowed because we'd be extending.
569 if (create && (iblock + max_blocks) > inode_blocks) {
574 /* This figures out the size of the next contiguous block, and
575 * our logical offset */
576 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
577 &contig_blocks, &ext_flags);
579 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
580 (unsigned long long)iblock);
585 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno && create) {
586 ocfs2_error(inode->i_sb,
587 "Inode %llu has a hole at block %llu\n",
588 (unsigned long long)OCFS2_I(inode)->ip_blkno,
589 (unsigned long long)iblock);
594 /* We should already CoW the refcounted extent in case of create. */
595 BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
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 /* Fallback to buffered I/O if we are appending. */
695 if (i_size_read(inode) <= offset)
698 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
699 inode->i_sb->s_bdev, iov, offset,
701 ocfs2_direct_IO_get_blocks,
708 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
713 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
715 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
718 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
720 cluster_start = cpos % cpp;
721 cluster_start = cluster_start << osb->s_clustersize_bits;
723 cluster_end = cluster_start + osb->s_clustersize;
726 BUG_ON(cluster_start > PAGE_SIZE);
727 BUG_ON(cluster_end > PAGE_SIZE);
730 *start = cluster_start;
736 * 'from' and 'to' are the region in the page to avoid zeroing.
738 * If pagesize > clustersize, this function will avoid zeroing outside
739 * of the cluster boundary.
741 * from == to == 0 is code for "zero the entire cluster region"
743 static void ocfs2_clear_page_regions(struct page *page,
744 struct ocfs2_super *osb, u32 cpos,
745 unsigned from, unsigned to)
748 unsigned int cluster_start, cluster_end;
750 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
752 kaddr = kmap_atomic(page, KM_USER0);
755 if (from > cluster_start)
756 memset(kaddr + cluster_start, 0, from - cluster_start);
757 if (to < cluster_end)
758 memset(kaddr + to, 0, cluster_end - to);
760 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
763 kunmap_atomic(kaddr, KM_USER0);
767 * Nonsparse file systems fully allocate before we get to the write
768 * code. This prevents ocfs2_write() from tagging the write as an
769 * allocating one, which means ocfs2_map_page_blocks() might try to
770 * read-in the blocks at the tail of our file. Avoid reading them by
771 * testing i_size against each block offset.
773 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
774 unsigned int block_start)
776 u64 offset = page_offset(page) + block_start;
778 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
781 if (i_size_read(inode) > offset)
788 * Some of this taken from block_prepare_write(). We already have our
789 * mapping by now though, and the entire write will be allocating or
790 * it won't, so not much need to use BH_New.
792 * This will also skip zeroing, which is handled externally.
794 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
795 struct inode *inode, unsigned int from,
796 unsigned int to, int new)
799 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
800 unsigned int block_end, block_start;
801 unsigned int bsize = 1 << inode->i_blkbits;
803 if (!page_has_buffers(page))
804 create_empty_buffers(page, bsize, 0);
806 head = page_buffers(page);
807 for (bh = head, block_start = 0; bh != head || !block_start;
808 bh = bh->b_this_page, block_start += bsize) {
809 block_end = block_start + bsize;
811 clear_buffer_new(bh);
814 * Ignore blocks outside of our i/o range -
815 * they may belong to unallocated clusters.
817 if (block_start >= to || block_end <= from) {
818 if (PageUptodate(page))
819 set_buffer_uptodate(bh);
824 * For an allocating write with cluster size >= page
825 * size, we always write the entire page.
830 if (!buffer_mapped(bh)) {
831 map_bh(bh, inode->i_sb, *p_blkno);
832 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
835 if (PageUptodate(page)) {
836 if (!buffer_uptodate(bh))
837 set_buffer_uptodate(bh);
838 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
840 ocfs2_should_read_blk(inode, page, block_start) &&
841 (block_start < from || block_end > to)) {
842 ll_rw_block(READ, 1, &bh);
846 *p_blkno = *p_blkno + 1;
850 * If we issued read requests - let them complete.
852 while(wait_bh > wait) {
853 wait_on_buffer(*--wait_bh);
854 if (!buffer_uptodate(*wait_bh))
858 if (ret == 0 || !new)
862 * If we get -EIO above, zero out any newly allocated blocks
863 * to avoid exposing stale data.
868 block_end = block_start + bsize;
869 if (block_end <= from)
871 if (block_start >= to)
874 zero_user(page, block_start, bh->b_size);
875 set_buffer_uptodate(bh);
876 mark_buffer_dirty(bh);
879 block_start = block_end;
880 bh = bh->b_this_page;
881 } while (bh != head);
886 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
887 #define OCFS2_MAX_CTXT_PAGES 1
889 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
892 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
895 * Describe the state of a single cluster to be written to.
897 struct ocfs2_write_cluster_desc {
901 * Give this a unique field because c_phys eventually gets
905 unsigned c_unwritten;
906 unsigned c_needs_zero;
909 struct ocfs2_write_ctxt {
910 /* Logical cluster position / len of write */
914 /* First cluster allocated in a nonsparse extend */
915 u32 w_first_new_cpos;
917 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
920 * This is true if page_size > cluster_size.
922 * It triggers a set of special cases during write which might
923 * have to deal with allocating writes to partial pages.
925 unsigned int w_large_pages;
928 * Pages involved in this write.
930 * w_target_page is the page being written to by the user.
932 * w_pages is an array of pages which always contains
933 * w_target_page, and in the case of an allocating write with
934 * page_size < cluster size, it will contain zero'd and mapped
935 * pages adjacent to w_target_page which need to be written
936 * out in so that future reads from that region will get
939 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
940 unsigned int w_num_pages;
941 struct page *w_target_page;
944 * ocfs2_write_end() uses this to know what the real range to
945 * write in the target should be.
947 unsigned int w_target_from;
948 unsigned int w_target_to;
951 * We could use journal_current_handle() but this is cleaner,
956 struct buffer_head *w_di_bh;
958 struct ocfs2_cached_dealloc_ctxt w_dealloc;
961 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
965 for(i = 0; i < num_pages; i++) {
967 unlock_page(pages[i]);
968 mark_page_accessed(pages[i]);
969 page_cache_release(pages[i]);
974 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
976 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
982 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
983 struct ocfs2_super *osb, loff_t pos,
984 unsigned len, struct buffer_head *di_bh)
987 struct ocfs2_write_ctxt *wc;
989 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
993 wc->w_cpos = pos >> osb->s_clustersize_bits;
994 wc->w_first_new_cpos = UINT_MAX;
995 cend = (pos + len - 1) >> osb->s_clustersize_bits;
996 wc->w_clen = cend - wc->w_cpos + 1;
1000 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
1001 wc->w_large_pages = 1;
1003 wc->w_large_pages = 0;
1005 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1013 * If a page has any new buffers, zero them out here, and mark them uptodate
1014 * and dirty so they'll be written out (in order to prevent uninitialised
1015 * block data from leaking). And clear the new bit.
1017 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1019 unsigned int block_start, block_end;
1020 struct buffer_head *head, *bh;
1022 BUG_ON(!PageLocked(page));
1023 if (!page_has_buffers(page))
1026 bh = head = page_buffers(page);
1029 block_end = block_start + bh->b_size;
1031 if (buffer_new(bh)) {
1032 if (block_end > from && block_start < to) {
1033 if (!PageUptodate(page)) {
1034 unsigned start, end;
1036 start = max(from, block_start);
1037 end = min(to, block_end);
1039 zero_user_segment(page, start, end);
1040 set_buffer_uptodate(bh);
1043 clear_buffer_new(bh);
1044 mark_buffer_dirty(bh);
1048 block_start = block_end;
1049 bh = bh->b_this_page;
1050 } while (bh != head);
1054 * Only called when we have a failure during allocating write to write
1055 * zero's to the newly allocated region.
1057 static void ocfs2_write_failure(struct inode *inode,
1058 struct ocfs2_write_ctxt *wc,
1059 loff_t user_pos, unsigned user_len)
1062 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1063 to = user_pos + user_len;
1064 struct page *tmppage;
1066 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1068 for(i = 0; i < wc->w_num_pages; i++) {
1069 tmppage = wc->w_pages[i];
1071 if (page_has_buffers(tmppage)) {
1072 if (ocfs2_should_order_data(inode))
1073 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1075 block_commit_write(tmppage, from, to);
1080 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1081 struct ocfs2_write_ctxt *wc,
1082 struct page *page, u32 cpos,
1083 loff_t user_pos, unsigned user_len,
1087 unsigned int map_from = 0, map_to = 0;
1088 unsigned int cluster_start, cluster_end;
1089 unsigned int user_data_from = 0, user_data_to = 0;
1091 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1092 &cluster_start, &cluster_end);
1094 /* treat the write as new if the a hole/lseek spanned across
1095 * the page boundary.
1097 new = new | ((i_size_read(inode) <= page_offset(page)) &&
1098 (page_offset(page) <= user_pos));
1100 if (page == wc->w_target_page) {
1101 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1102 map_to = map_from + user_len;
1105 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1106 cluster_start, cluster_end,
1109 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1110 map_from, map_to, new);
1116 user_data_from = map_from;
1117 user_data_to = map_to;
1119 map_from = cluster_start;
1120 map_to = cluster_end;
1124 * If we haven't allocated the new page yet, we
1125 * shouldn't be writing it out without copying user
1126 * data. This is likely a math error from the caller.
1130 map_from = cluster_start;
1131 map_to = cluster_end;
1133 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1134 cluster_start, cluster_end, new);
1142 * Parts of newly allocated pages need to be zero'd.
1144 * Above, we have also rewritten 'to' and 'from' - as far as
1145 * the rest of the function is concerned, the entire cluster
1146 * range inside of a page needs to be written.
1148 * We can skip this if the page is up to date - it's already
1149 * been zero'd from being read in as a hole.
1151 if (new && !PageUptodate(page))
1152 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1153 cpos, user_data_from, user_data_to);
1155 flush_dcache_page(page);
1162 * This function will only grab one clusters worth of pages.
1164 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1165 struct ocfs2_write_ctxt *wc,
1166 u32 cpos, loff_t user_pos, int new,
1167 struct page *mmap_page)
1170 unsigned long start, target_index, index;
1171 struct inode *inode = mapping->host;
1173 target_index = user_pos >> PAGE_CACHE_SHIFT;
1176 * Figure out how many pages we'll be manipulating here. For
1177 * non allocating write, we just change the one
1178 * page. Otherwise, we'll need a whole clusters worth.
1181 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1182 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1184 wc->w_num_pages = 1;
1185 start = target_index;
1188 for(i = 0; i < wc->w_num_pages; i++) {
1191 if (index == target_index && mmap_page) {
1193 * ocfs2_pagemkwrite() is a little different
1194 * and wants us to directly use the page
1197 lock_page(mmap_page);
1199 if (mmap_page->mapping != mapping) {
1200 unlock_page(mmap_page);
1202 * Sanity check - the locking in
1203 * ocfs2_pagemkwrite() should ensure
1204 * that this code doesn't trigger.
1211 page_cache_get(mmap_page);
1212 wc->w_pages[i] = mmap_page;
1214 wc->w_pages[i] = find_or_create_page(mapping, index,
1216 if (!wc->w_pages[i]) {
1223 if (index == target_index)
1224 wc->w_target_page = wc->w_pages[i];
1231 * Prepare a single cluster for write one cluster into the file.
1233 static int ocfs2_write_cluster(struct address_space *mapping,
1234 u32 phys, unsigned int unwritten,
1235 unsigned int should_zero,
1236 struct ocfs2_alloc_context *data_ac,
1237 struct ocfs2_alloc_context *meta_ac,
1238 struct ocfs2_write_ctxt *wc, u32 cpos,
1239 loff_t user_pos, unsigned user_len)
1242 u64 v_blkno, p_blkno;
1243 struct inode *inode = mapping->host;
1244 struct ocfs2_extent_tree et;
1246 new = phys == 0 ? 1 : 0;
1251 * This is safe to call with the page locks - it won't take
1252 * any additional semaphores or cluster locks.
1255 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1256 &tmp_pos, 1, 0, wc->w_di_bh,
1257 wc->w_handle, data_ac,
1260 * This shouldn't happen because we must have already
1261 * calculated the correct meta data allocation required. The
1262 * internal tree allocation code should know how to increase
1263 * transaction credits itself.
1265 * If need be, we could handle -EAGAIN for a
1266 * RESTART_TRANS here.
1268 mlog_bug_on_msg(ret == -EAGAIN,
1269 "Inode %llu: EAGAIN return during allocation.\n",
1270 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1275 } else if (unwritten) {
1276 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1278 ret = ocfs2_mark_extent_written(inode, &et,
1279 wc->w_handle, cpos, 1, phys,
1280 meta_ac, &wc->w_dealloc);
1288 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1290 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1293 * The only reason this should fail is due to an inability to
1294 * find the extent added.
1296 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1299 ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, "
1300 "at logical block %llu",
1301 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1302 (unsigned long long)v_blkno);
1306 BUG_ON(p_blkno == 0);
1308 for(i = 0; i < wc->w_num_pages; i++) {
1311 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1312 wc->w_pages[i], cpos,
1323 * We only have cleanup to do in case of allocating write.
1326 ocfs2_write_failure(inode, wc, user_pos, user_len);
1333 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1334 struct ocfs2_alloc_context *data_ac,
1335 struct ocfs2_alloc_context *meta_ac,
1336 struct ocfs2_write_ctxt *wc,
1337 loff_t pos, unsigned len)
1341 unsigned int local_len = len;
1342 struct ocfs2_write_cluster_desc *desc;
1343 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1345 for (i = 0; i < wc->w_clen; i++) {
1346 desc = &wc->w_desc[i];
1349 * We have to make sure that the total write passed in
1350 * doesn't extend past a single cluster.
1353 cluster_off = pos & (osb->s_clustersize - 1);
1354 if ((cluster_off + local_len) > osb->s_clustersize)
1355 local_len = osb->s_clustersize - cluster_off;
1357 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1361 wc, desc->c_cpos, pos, local_len);
1377 * ocfs2_write_end() wants to know which parts of the target page it
1378 * should complete the write on. It's easiest to compute them ahead of
1379 * time when a more complete view of the write is available.
1381 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1382 struct ocfs2_write_ctxt *wc,
1383 loff_t pos, unsigned len, int alloc)
1385 struct ocfs2_write_cluster_desc *desc;
1387 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1388 wc->w_target_to = wc->w_target_from + len;
1394 * Allocating write - we may have different boundaries based
1395 * on page size and cluster size.
1397 * NOTE: We can no longer compute one value from the other as
1398 * the actual write length and user provided length may be
1402 if (wc->w_large_pages) {
1404 * We only care about the 1st and last cluster within
1405 * our range and whether they should be zero'd or not. Either
1406 * value may be extended out to the start/end of a
1407 * newly allocated cluster.
1409 desc = &wc->w_desc[0];
1410 if (desc->c_needs_zero)
1411 ocfs2_figure_cluster_boundaries(osb,
1416 desc = &wc->w_desc[wc->w_clen - 1];
1417 if (desc->c_needs_zero)
1418 ocfs2_figure_cluster_boundaries(osb,
1423 wc->w_target_from = 0;
1424 wc->w_target_to = PAGE_CACHE_SIZE;
1429 * Populate each single-cluster write descriptor in the write context
1430 * with information about the i/o to be done.
1432 * Returns the number of clusters that will have to be allocated, as
1433 * well as a worst case estimate of the number of extent records that
1434 * would have to be created during a write to an unwritten region.
1436 static int ocfs2_populate_write_desc(struct inode *inode,
1437 struct ocfs2_write_ctxt *wc,
1438 unsigned int *clusters_to_alloc,
1439 unsigned int *extents_to_split)
1442 struct ocfs2_write_cluster_desc *desc;
1443 unsigned int num_clusters = 0;
1444 unsigned int ext_flags = 0;
1448 *clusters_to_alloc = 0;
1449 *extents_to_split = 0;
1451 for (i = 0; i < wc->w_clen; i++) {
1452 desc = &wc->w_desc[i];
1453 desc->c_cpos = wc->w_cpos + i;
1455 if (num_clusters == 0) {
1457 * Need to look up the next extent record.
1459 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1460 &num_clusters, &ext_flags);
1466 /* We should already CoW the refcountd extent. */
1467 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1470 * Assume worst case - that we're writing in
1471 * the middle of the extent.
1473 * We can assume that the write proceeds from
1474 * left to right, in which case the extent
1475 * insert code is smart enough to coalesce the
1476 * next splits into the previous records created.
1478 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1479 *extents_to_split = *extents_to_split + 2;
1482 * Only increment phys if it doesn't describe
1489 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1490 * file that got extended. w_first_new_cpos tells us
1491 * where the newly allocated clusters are so we can
1494 if (desc->c_cpos >= wc->w_first_new_cpos) {
1496 desc->c_needs_zero = 1;
1499 desc->c_phys = phys;
1502 desc->c_needs_zero = 1;
1503 *clusters_to_alloc = *clusters_to_alloc + 1;
1506 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1507 desc->c_unwritten = 1;
1508 desc->c_needs_zero = 1;
1519 static int ocfs2_write_begin_inline(struct address_space *mapping,
1520 struct inode *inode,
1521 struct ocfs2_write_ctxt *wc)
1524 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1527 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1529 page = find_or_create_page(mapping, 0, GFP_NOFS);
1536 * If we don't set w_num_pages then this page won't get unlocked
1537 * and freed on cleanup of the write context.
1539 wc->w_pages[0] = wc->w_target_page = page;
1540 wc->w_num_pages = 1;
1542 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1543 if (IS_ERR(handle)) {
1544 ret = PTR_ERR(handle);
1549 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1550 OCFS2_JOURNAL_ACCESS_WRITE);
1552 ocfs2_commit_trans(osb, handle);
1558 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1559 ocfs2_set_inode_data_inline(inode, di);
1561 if (!PageUptodate(page)) {
1562 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1564 ocfs2_commit_trans(osb, handle);
1570 wc->w_handle = handle;
1575 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1577 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1579 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1584 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1585 struct inode *inode, loff_t pos,
1586 unsigned len, struct page *mmap_page,
1587 struct ocfs2_write_ctxt *wc)
1589 int ret, written = 0;
1590 loff_t end = pos + len;
1591 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1592 struct ocfs2_dinode *di = NULL;
1594 mlog(0, "Inode %llu, write of %u bytes at off %llu. features: 0x%x\n",
1595 (unsigned long long)oi->ip_blkno, len, (unsigned long long)pos,
1596 oi->ip_dyn_features);
1599 * Handle inodes which already have inline data 1st.
1601 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1602 if (mmap_page == NULL &&
1603 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1604 goto do_inline_write;
1607 * The write won't fit - we have to give this inode an
1608 * inline extent list now.
1610 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1617 * Check whether the inode can accept inline data.
1619 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1623 * Check whether the write can fit.
1625 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1627 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1631 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1638 * This signals to the caller that the data can be written
1643 return written ? written : ret;
1647 * This function only does anything for file systems which can't
1648 * handle sparse files.
1650 * What we want to do here is fill in any hole between the current end
1651 * of allocation and the end of our write. That way the rest of the
1652 * write path can treat it as an non-allocating write, which has no
1653 * special case code for sparse/nonsparse files.
1655 static int ocfs2_expand_nonsparse_inode(struct inode *inode, loff_t pos,
1657 struct ocfs2_write_ctxt *wc)
1660 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1661 loff_t newsize = pos + len;
1663 if (ocfs2_sparse_alloc(osb))
1666 if (newsize <= i_size_read(inode))
1669 ret = ocfs2_extend_no_holes(inode, newsize, pos);
1673 wc->w_first_new_cpos =
1674 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1679 int ocfs2_write_begin_nolock(struct address_space *mapping,
1680 loff_t pos, unsigned len, unsigned flags,
1681 struct page **pagep, void **fsdata,
1682 struct buffer_head *di_bh, struct page *mmap_page)
1684 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1685 unsigned int clusters_to_alloc, extents_to_split;
1686 struct ocfs2_write_ctxt *wc;
1687 struct inode *inode = mapping->host;
1688 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1689 struct ocfs2_dinode *di;
1690 struct ocfs2_alloc_context *data_ac = NULL;
1691 struct ocfs2_alloc_context *meta_ac = NULL;
1693 struct ocfs2_extent_tree et;
1695 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1701 if (ocfs2_supports_inline_data(osb)) {
1702 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1714 ret = ocfs2_expand_nonsparse_inode(inode, pos, len, wc);
1720 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1724 } else if (ret == 1) {
1725 ret = ocfs2_refcount_cow(inode, di_bh,
1726 wc->w_cpos, wc->w_clen, UINT_MAX);
1733 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1740 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1743 * We set w_target_from, w_target_to here so that
1744 * ocfs2_write_end() knows which range in the target page to
1745 * write out. An allocation requires that we write the entire
1748 if (clusters_to_alloc || extents_to_split) {
1750 * XXX: We are stretching the limits of
1751 * ocfs2_lock_allocators(). It greatly over-estimates
1752 * the work to be done.
1754 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u,"
1755 " clusters_to_add = %u, extents_to_split = %u\n",
1756 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1757 (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
1758 clusters_to_alloc, extents_to_split);
1760 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1762 ret = ocfs2_lock_allocators(inode, &et,
1763 clusters_to_alloc, extents_to_split,
1764 &data_ac, &meta_ac);
1770 credits = ocfs2_calc_extend_credits(inode->i_sb,
1777 * We have to zero sparse allocated clusters, unwritten extent clusters,
1778 * and non-sparse clusters we just extended. For non-sparse writes,
1779 * we know zeros will only be needed in the first and/or last cluster.
1781 if (clusters_to_alloc || extents_to_split ||
1782 (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1783 wc->w_desc[wc->w_clen - 1].c_needs_zero)))
1784 cluster_of_pages = 1;
1786 cluster_of_pages = 0;
1788 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
1790 handle = ocfs2_start_trans(osb, credits);
1791 if (IS_ERR(handle)) {
1792 ret = PTR_ERR(handle);
1797 wc->w_handle = handle;
1799 if (clusters_to_alloc && vfs_dq_alloc_space_nodirty(inode,
1800 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc))) {
1805 * We don't want this to fail in ocfs2_write_end(), so do it
1808 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1809 OCFS2_JOURNAL_ACCESS_WRITE);
1816 * Fill our page array first. That way we've grabbed enough so
1817 * that we can zero and flush if we error after adding the
1820 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
1821 cluster_of_pages, mmap_page);
1827 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
1835 ocfs2_free_alloc_context(data_ac);
1837 ocfs2_free_alloc_context(meta_ac);
1840 *pagep = wc->w_target_page;
1844 if (clusters_to_alloc)
1845 vfs_dq_free_space(inode,
1846 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
1848 ocfs2_commit_trans(osb, handle);
1851 ocfs2_free_write_ctxt(wc);
1854 ocfs2_free_alloc_context(data_ac);
1856 ocfs2_free_alloc_context(meta_ac);
1860 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
1861 loff_t pos, unsigned len, unsigned flags,
1862 struct page **pagep, void **fsdata)
1865 struct buffer_head *di_bh = NULL;
1866 struct inode *inode = mapping->host;
1868 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1875 * Take alloc sem here to prevent concurrent lookups. That way
1876 * the mapping, zeroing and tree manipulation within
1877 * ocfs2_write() will be safe against ->readpage(). This
1878 * should also serve to lock out allocation from a shared
1881 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1883 ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep,
1884 fsdata, di_bh, NULL);
1895 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1898 ocfs2_inode_unlock(inode, 1);
1903 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
1904 unsigned len, unsigned *copied,
1905 struct ocfs2_dinode *di,
1906 struct ocfs2_write_ctxt *wc)
1910 if (unlikely(*copied < len)) {
1911 if (!PageUptodate(wc->w_target_page)) {
1917 kaddr = kmap_atomic(wc->w_target_page, KM_USER0);
1918 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
1919 kunmap_atomic(kaddr, KM_USER0);
1921 mlog(0, "Data written to inode at offset %llu. "
1922 "id_count = %u, copied = %u, i_dyn_features = 0x%x\n",
1923 (unsigned long long)pos, *copied,
1924 le16_to_cpu(di->id2.i_data.id_count),
1925 le16_to_cpu(di->i_dyn_features));
1928 int ocfs2_write_end_nolock(struct address_space *mapping,
1929 loff_t pos, unsigned len, unsigned copied,
1930 struct page *page, void *fsdata)
1933 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
1934 struct inode *inode = mapping->host;
1935 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1936 struct ocfs2_write_ctxt *wc = fsdata;
1937 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1938 handle_t *handle = wc->w_handle;
1939 struct page *tmppage;
1941 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1942 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
1943 goto out_write_size;
1946 if (unlikely(copied < len)) {
1947 if (!PageUptodate(wc->w_target_page))
1950 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
1953 flush_dcache_page(wc->w_target_page);
1955 for(i = 0; i < wc->w_num_pages; i++) {
1956 tmppage = wc->w_pages[i];
1958 if (tmppage == wc->w_target_page) {
1959 from = wc->w_target_from;
1960 to = wc->w_target_to;
1962 BUG_ON(from > PAGE_CACHE_SIZE ||
1963 to > PAGE_CACHE_SIZE ||
1967 * Pages adjacent to the target (if any) imply
1968 * a hole-filling write in which case we want
1969 * to flush their entire range.
1972 to = PAGE_CACHE_SIZE;
1975 if (page_has_buffers(tmppage)) {
1976 if (ocfs2_should_order_data(inode))
1977 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1978 block_commit_write(tmppage, from, to);
1984 if (pos > inode->i_size) {
1985 i_size_write(inode, pos);
1986 mark_inode_dirty(inode);
1988 inode->i_blocks = ocfs2_inode_sector_count(inode);
1989 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1990 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1991 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1992 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1993 ocfs2_journal_dirty(handle, wc->w_di_bh);
1995 ocfs2_commit_trans(osb, handle);
1997 ocfs2_run_deallocs(osb, &wc->w_dealloc);
1999 ocfs2_free_write_ctxt(wc);
2004 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2005 loff_t pos, unsigned len, unsigned copied,
2006 struct page *page, void *fsdata)
2009 struct inode *inode = mapping->host;
2011 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
2013 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2014 ocfs2_inode_unlock(inode, 1);
2019 const struct address_space_operations ocfs2_aops = {
2020 .readpage = ocfs2_readpage,
2021 .readpages = ocfs2_readpages,
2022 .writepage = ocfs2_writepage,
2023 .write_begin = ocfs2_write_begin,
2024 .write_end = ocfs2_write_end,
2026 .sync_page = block_sync_page,
2027 .direct_IO = ocfs2_direct_IO,
2028 .invalidatepage = ocfs2_invalidatepage,
2029 .releasepage = ocfs2_releasepage,
2030 .migratepage = buffer_migrate_page,
2031 .is_partially_uptodate = block_is_partially_uptodate,
2032 .error_remove_page = generic_error_remove_page,