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>
30 #define MLOG_MASK_PREFIX ML_FILE_IO
31 #include <cluster/masklog.h>
38 #include "extent_map.h"
46 #include "buffer_head_io.h"
48 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
49 struct buffer_head *bh_result, int create)
53 struct ocfs2_dinode *fe = NULL;
54 struct buffer_head *bh = NULL;
55 struct buffer_head *buffer_cache_bh = NULL;
56 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
59 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
60 (unsigned long long)iblock, bh_result, create);
62 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
64 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
65 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
66 (unsigned long long)iblock);
70 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
71 OCFS2_I(inode)->ip_blkno,
72 &bh, OCFS2_BH_CACHED, inode);
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)fe->i_blkno, 7, fe->i_signature);
85 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
86 le32_to_cpu(fe->i_clusters))) {
87 mlog(ML_ERROR, "block offset is outside the allocated size: "
88 "%llu\n", (unsigned long long)iblock);
92 /* We don't use the page cache to create symlink data, so if
93 * need be, copy it over from the buffer cache. */
94 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
95 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
97 buffer_cache_bh = sb_getblk(osb->sb, blkno);
98 if (!buffer_cache_bh) {
99 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
103 /* we haven't locked out transactions, so a commit
104 * could've happened. Since we've got a reference on
105 * the bh, even if it commits while we're doing the
106 * copy, the data is still good. */
107 if (buffer_jbd(buffer_cache_bh)
108 && ocfs2_inode_is_new(inode)) {
109 kaddr = kmap_atomic(bh_result->b_page, KM_USER0);
111 mlog(ML_ERROR, "couldn't kmap!\n");
114 memcpy(kaddr + (bh_result->b_size * iblock),
115 buffer_cache_bh->b_data,
117 kunmap_atomic(kaddr, KM_USER0);
118 set_buffer_uptodate(bh_result);
120 brelse(buffer_cache_bh);
123 map_bh(bh_result, inode->i_sb,
124 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 u64 p_blkno, past_eof;
141 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
143 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
144 (unsigned long long)iblock, bh_result, create);
146 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
147 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
148 inode, inode->i_ino);
150 if (S_ISLNK(inode->i_mode)) {
151 /* this always does I/O for some reason. */
152 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
156 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL);
158 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
159 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
160 (unsigned long long)p_blkno);
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 mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb),
171 "ino %lu, iblock %llu\n", inode->i_ino,
172 (unsigned long long)iblock);
175 map_bh(bh_result, inode->i_sb, p_blkno);
177 if (!ocfs2_sparse_alloc(osb)) {
181 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
182 (unsigned long long)iblock,
183 (unsigned long long)p_blkno,
184 (unsigned long long)OCFS2_I(inode)->ip_blkno);
185 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
189 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
190 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
191 (unsigned long long)past_eof);
193 if (create && (iblock >= past_eof))
194 set_buffer_new(bh_result);
205 static int ocfs2_readpage(struct file *file, struct page *page)
207 struct inode *inode = page->mapping->host;
208 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
211 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
213 ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page);
215 if (ret == AOP_TRUNCATED_PAGE)
221 down_read(&OCFS2_I(inode)->ip_alloc_sem);
224 * i_size might have just been updated as we grabed the meta lock. We
225 * might now be discovering a truncate that hit on another node.
226 * block_read_full_page->get_block freaks out if it is asked to read
227 * beyond the end of a file, so we check here. Callers
228 * (generic_file_read, fault->nopage) are clever enough to check i_size
229 * and notice that the page they just read isn't needed.
231 * XXX sys_readahead() seems to get that wrong?
233 if (start >= i_size_read(inode)) {
234 char *addr = kmap(page);
235 memset(addr, 0, PAGE_SIZE);
236 flush_dcache_page(page);
238 SetPageUptodate(page);
243 ret = ocfs2_data_lock_with_page(inode, 0, page);
245 if (ret == AOP_TRUNCATED_PAGE)
251 ret = block_read_full_page(page, ocfs2_get_block);
254 ocfs2_data_unlock(inode, 0);
256 up_read(&OCFS2_I(inode)->ip_alloc_sem);
257 ocfs2_meta_unlock(inode, 0);
265 /* Note: Because we don't support holes, our allocation has
266 * already happened (allocation writes zeros to the file data)
267 * so we don't have to worry about ordered writes in
270 * ->writepage is called during the process of invalidating the page cache
271 * during blocked lock processing. It can't block on any cluster locks
272 * to during block mapping. It's relying on the fact that the block
273 * mapping can't have disappeared under the dirty pages that it is
274 * being asked to write back.
276 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
280 mlog_entry("(0x%p)\n", page);
282 ret = block_write_full_page(page, ocfs2_get_block, wbc);
290 * This is called from ocfs2_write_zero_page() which has handled it's
291 * own cluster locking and has ensured allocation exists for those
292 * blocks to be written.
294 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
295 unsigned from, unsigned to)
299 down_read(&OCFS2_I(inode)->ip_alloc_sem);
301 ret = block_prepare_write(page, from, to, ocfs2_get_block);
303 up_read(&OCFS2_I(inode)->ip_alloc_sem);
308 /* Taken from ext3. We don't necessarily need the full blown
309 * functionality yet, but IMHO it's better to cut and paste the whole
310 * thing so we can avoid introducing our own bugs (and easily pick up
311 * their fixes when they happen) --Mark */
312 int walk_page_buffers( handle_t *handle,
313 struct buffer_head *head,
317 int (*fn)( handle_t *handle,
318 struct buffer_head *bh))
320 struct buffer_head *bh;
321 unsigned block_start, block_end;
322 unsigned blocksize = head->b_size;
324 struct buffer_head *next;
326 for ( bh = head, block_start = 0;
327 ret == 0 && (bh != head || !block_start);
328 block_start = block_end, bh = next)
330 next = bh->b_this_page;
331 block_end = block_start + blocksize;
332 if (block_end <= from || block_start >= to) {
333 if (partial && !buffer_uptodate(bh))
337 err = (*fn)(handle, bh);
344 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
349 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
350 handle_t *handle = NULL;
353 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
360 if (ocfs2_should_order_data(inode)) {
361 ret = walk_page_buffers(handle,
364 ocfs2_journal_dirty_data);
371 ocfs2_commit_trans(osb, handle);
372 handle = ERR_PTR(ret);
377 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
382 struct inode *inode = mapping->host;
384 mlog_entry("(block = %llu)\n", (unsigned long long)block);
386 /* We don't need to lock journal system files, since they aren't
387 * accessed concurrently from multiple nodes.
389 if (!INODE_JOURNAL(inode)) {
390 err = ocfs2_meta_lock(inode, NULL, 0);
396 down_read(&OCFS2_I(inode)->ip_alloc_sem);
399 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL);
401 if (!INODE_JOURNAL(inode)) {
402 up_read(&OCFS2_I(inode)->ip_alloc_sem);
403 ocfs2_meta_unlock(inode, 0);
407 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
408 (unsigned long long)block);
415 status = err ? 0 : p_blkno;
417 mlog_exit((int)status);
423 * TODO: Make this into a generic get_blocks function.
425 * From do_direct_io in direct-io.c:
426 * "So what we do is to permit the ->get_blocks function to populate
427 * bh.b_size with the size of IO which is permitted at this offset and
430 * This function is called directly from get_more_blocks in direct-io.c.
432 * called like this: dio->get_blocks(dio->inode, fs_startblk,
433 * fs_count, map_bh, dio->rw == WRITE);
435 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
436 struct buffer_head *bh_result, int create)
439 u64 p_blkno, inode_blocks;
441 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
442 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
444 /* This function won't even be called if the request isn't all
445 * nicely aligned and of the right size, so there's no need
446 * for us to check any of that. */
448 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
451 * Any write past EOF is not allowed because we'd be extending.
453 if (create && (iblock + max_blocks) > inode_blocks) {
458 /* This figures out the size of the next contiguous block, and
459 * our logical offset */
460 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
463 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
464 (unsigned long long)iblock);
469 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) {
470 ocfs2_error(inode->i_sb,
471 "Inode %llu has a hole at block %llu\n",
472 (unsigned long long)OCFS2_I(inode)->ip_blkno,
473 (unsigned long long)iblock);
479 * get_more_blocks() expects us to describe a hole by clearing
480 * the mapped bit on bh_result().
483 map_bh(bh_result, inode->i_sb, p_blkno);
486 * ocfs2_prepare_inode_for_write() should have caught
487 * the case where we'd be filling a hole and triggered
488 * a buffered write instead.
496 clear_buffer_mapped(bh_result);
499 /* make sure we don't map more than max_blocks blocks here as
500 that's all the kernel will handle at this point. */
501 if (max_blocks < contig_blocks)
502 contig_blocks = max_blocks;
503 bh_result->b_size = contig_blocks << blocksize_bits;
509 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
510 * particularly interested in the aio/dio case. Like the core uses
511 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
512 * truncation on another.
514 static void ocfs2_dio_end_io(struct kiocb *iocb,
519 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
521 /* this io's submitter should not have unlocked this before we could */
522 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
523 ocfs2_iocb_clear_rw_locked(iocb);
524 up_read(&inode->i_alloc_sem);
525 ocfs2_rw_unlock(inode, 0);
529 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
530 * from ext3. PageChecked() bits have been removed as OCFS2 does not
531 * do journalled data.
533 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
535 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
537 journal_invalidatepage(journal, page, offset);
540 static int ocfs2_releasepage(struct page *page, gfp_t wait)
542 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
544 if (!page_has_buffers(page))
546 return journal_try_to_free_buffers(journal, page, wait);
549 static ssize_t ocfs2_direct_IO(int rw,
551 const struct iovec *iov,
553 unsigned long nr_segs)
555 struct file *file = iocb->ki_filp;
556 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
561 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
563 * We get PR data locks even for O_DIRECT. This
564 * allows concurrent O_DIRECT I/O but doesn't let
565 * O_DIRECT with extending and buffered zeroing writes
566 * race. If they did race then the buffered zeroing
567 * could be written back after the O_DIRECT I/O. It's
568 * one thing to tell people not to mix buffered and
569 * O_DIRECT writes, but expecting them to understand
570 * that file extension is also an implicit buffered
571 * write is too much. By getting the PR we force
572 * writeback of the buffered zeroing before
575 ret = ocfs2_data_lock(inode, 0);
580 ocfs2_data_unlock(inode, 0);
583 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
584 inode->i_sb->s_bdev, iov, offset,
586 ocfs2_direct_IO_get_blocks,
593 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
598 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
600 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
603 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
605 cluster_start = cpos % cpp;
606 cluster_start = cluster_start << osb->s_clustersize_bits;
608 cluster_end = cluster_start + osb->s_clustersize;
611 BUG_ON(cluster_start > PAGE_SIZE);
612 BUG_ON(cluster_end > PAGE_SIZE);
615 *start = cluster_start;
621 * 'from' and 'to' are the region in the page to avoid zeroing.
623 * If pagesize > clustersize, this function will avoid zeroing outside
624 * of the cluster boundary.
626 * from == to == 0 is code for "zero the entire cluster region"
628 static void ocfs2_clear_page_regions(struct page *page,
629 struct ocfs2_super *osb, u32 cpos,
630 unsigned from, unsigned to)
633 unsigned int cluster_start, cluster_end;
635 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
637 kaddr = kmap_atomic(page, KM_USER0);
640 if (from > cluster_start)
641 memset(kaddr + cluster_start, 0, from - cluster_start);
642 if (to < cluster_end)
643 memset(kaddr + to, 0, cluster_end - to);
645 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
648 kunmap_atomic(kaddr, KM_USER0);
652 * Some of this taken from block_prepare_write(). We already have our
653 * mapping by now though, and the entire write will be allocating or
654 * it won't, so not much need to use BH_New.
656 * This will also skip zeroing, which is handled externally.
658 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
659 struct inode *inode, unsigned int from,
660 unsigned int to, int new)
663 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
664 unsigned int block_end, block_start;
665 unsigned int bsize = 1 << inode->i_blkbits;
667 if (!page_has_buffers(page))
668 create_empty_buffers(page, bsize, 0);
670 head = page_buffers(page);
671 for (bh = head, block_start = 0; bh != head || !block_start;
672 bh = bh->b_this_page, block_start += bsize) {
673 block_end = block_start + bsize;
676 * Ignore blocks outside of our i/o range -
677 * they may belong to unallocated clusters.
679 if (block_start >= to || block_end <= from) {
680 if (PageUptodate(page))
681 set_buffer_uptodate(bh);
686 * For an allocating write with cluster size >= page
687 * size, we always write the entire page.
691 clear_buffer_new(bh);
693 if (!buffer_mapped(bh)) {
694 map_bh(bh, inode->i_sb, *p_blkno);
695 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
698 if (PageUptodate(page)) {
699 if (!buffer_uptodate(bh))
700 set_buffer_uptodate(bh);
701 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
702 (block_start < from || block_end > to)) {
703 ll_rw_block(READ, 1, &bh);
707 *p_blkno = *p_blkno + 1;
711 * If we issued read requests - let them complete.
713 while(wait_bh > wait) {
714 wait_on_buffer(*--wait_bh);
715 if (!buffer_uptodate(*wait_bh))
719 if (ret == 0 || !new)
723 * If we get -EIO above, zero out any newly allocated blocks
724 * to avoid exposing stale data.
731 block_end = block_start + bsize;
732 if (block_end <= from)
734 if (block_start >= to)
737 kaddr = kmap_atomic(page, KM_USER0);
738 memset(kaddr+block_start, 0, bh->b_size);
739 flush_dcache_page(page);
740 kunmap_atomic(kaddr, KM_USER0);
741 set_buffer_uptodate(bh);
742 mark_buffer_dirty(bh);
745 block_start = block_end;
746 bh = bh->b_this_page;
747 } while (bh != head);
753 * This will copy user data from the buffer page in the splice
756 * For now, we ignore SPLICE_F_MOVE as that would require some extra
757 * communication out all the way to ocfs2_write().
759 int ocfs2_map_and_write_splice_data(struct inode *inode,
760 struct ocfs2_write_ctxt *wc, u64 *p_blkno,
761 unsigned int *ret_from, unsigned int *ret_to)
764 unsigned int to, from, cluster_start, cluster_end;
766 struct ocfs2_splice_write_priv *sp = wc->w_private;
767 struct pipe_buffer *buf = sp->s_buf;
768 unsigned long bytes, src_from;
769 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
771 ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
775 src_from = sp->s_buf_offset;
778 if (wc->w_large_pages) {
780 * For cluster size < page size, we have to
781 * calculate pos within the cluster and obey
782 * the rightmost boundary.
784 bytes = min(bytes, (unsigned long)(osb->s_clustersize
785 - (wc->w_pos & (osb->s_clustersize - 1))));
789 if (wc->w_this_page_new)
790 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
791 cluster_start, cluster_end, 1);
793 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
800 BUG_ON(from > PAGE_CACHE_SIZE);
801 BUG_ON(to > PAGE_CACHE_SIZE);
802 BUG_ON(from > osb->s_clustersize);
803 BUG_ON(to > osb->s_clustersize);
805 src = buf->ops->map(sp->s_pipe, buf, 1);
806 dst = kmap_atomic(wc->w_this_page, KM_USER1);
807 memcpy(dst + from, src + src_from, bytes);
808 kunmap_atomic(wc->w_this_page, KM_USER1);
809 buf->ops->unmap(sp->s_pipe, buf, src);
811 wc->w_finished_copy = 1;
817 return bytes ? (unsigned int)bytes : ret;
821 * This will copy user data from the iovec in the buffered write
824 int ocfs2_map_and_write_user_data(struct inode *inode,
825 struct ocfs2_write_ctxt *wc, u64 *p_blkno,
826 unsigned int *ret_from, unsigned int *ret_to)
829 unsigned int to, from, cluster_start, cluster_end;
830 unsigned long bytes, src_from;
832 struct ocfs2_buffered_write_priv *bp = wc->w_private;
833 const struct iovec *cur_iov = bp->b_cur_iov;
835 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
837 ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
840 buf = cur_iov->iov_base + bp->b_cur_off;
841 src_from = (unsigned long)buf & ~PAGE_CACHE_MASK;
843 from = wc->w_pos & (PAGE_CACHE_SIZE - 1);
846 * This is a lot of comparisons, but it reads quite
847 * easily, which is important here.
849 /* Stay within the src page */
850 bytes = PAGE_SIZE - src_from;
851 /* Stay within the vector */
853 (unsigned long)(cur_iov->iov_len - bp->b_cur_off));
854 /* Stay within count */
855 bytes = min(bytes, (unsigned long)wc->w_count);
857 * For clustersize > page size, just stay within
858 * target page, otherwise we have to calculate pos
859 * within the cluster and obey the rightmost
862 if (wc->w_large_pages) {
864 * For cluster size < page size, we have to
865 * calculate pos within the cluster and obey
866 * the rightmost boundary.
868 bytes = min(bytes, (unsigned long)(osb->s_clustersize
869 - (wc->w_pos & (osb->s_clustersize - 1))));
872 * cluster size > page size is the most common
873 * case - we just stay within the target page
876 bytes = min(bytes, PAGE_CACHE_SIZE - from);
881 if (wc->w_this_page_new)
882 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
883 cluster_start, cluster_end, 1);
885 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
892 BUG_ON(from > PAGE_CACHE_SIZE);
893 BUG_ON(to > PAGE_CACHE_SIZE);
894 BUG_ON(from > osb->s_clustersize);
895 BUG_ON(to > osb->s_clustersize);
897 dst = kmap(wc->w_this_page);
898 memcpy(dst + from, bp->b_src_buf + src_from, bytes);
899 kunmap(wc->w_this_page);
902 * XXX: This is slow, but simple. The caller of
903 * ocfs2_buffered_write_cluster() is responsible for
904 * passing through the iovecs, so it's difficult to
905 * predict what our next step is in here after our
906 * initial write. A future version should be pushing
907 * that iovec manipulation further down.
909 * By setting this, we indicate that a copy from user
910 * data was done, and subsequent calls for this
911 * cluster will skip copying more data.
913 wc->w_finished_copy = 1;
919 return bytes ? (unsigned int)bytes : ret;
923 * Map, fill and write a page to disk.
925 * The work of copying data is done via callback. Newly allocated
926 * pages which don't take user data will be zero'd (set 'new' to
927 * indicate an allocating write)
929 * Returns a negative error code or the number of bytes copied into
932 int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
933 u64 *p_blkno, struct page *page,
934 struct ocfs2_write_ctxt *wc, int new)
937 unsigned int from = 0, to = 0;
938 unsigned int cluster_start, cluster_end;
939 unsigned int zero_from = 0, zero_to = 0;
941 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), wc->w_cpos,
942 &cluster_start, &cluster_end);
944 if ((wc->w_pos >> PAGE_CACHE_SHIFT) == page->index
945 && !wc->w_finished_copy) {
947 wc->w_this_page = page;
948 wc->w_this_page_new = new;
949 ret = wc->w_write_data_page(inode, wc, p_blkno, &from, &to);
960 from = cluster_start;
965 * If we haven't allocated the new page yet, we
966 * shouldn't be writing it out without copying user
967 * data. This is likely a math error from the caller.
971 from = cluster_start;
974 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
975 cluster_start, cluster_end, 1);
983 * Parts of newly allocated pages need to be zero'd.
985 * Above, we have also rewritten 'to' and 'from' - as far as
986 * the rest of the function is concerned, the entire cluster
987 * range inside of a page needs to be written.
989 * We can skip this if the page is up to date - it's already
990 * been zero'd from being read in as a hole.
992 if (new && !PageUptodate(page))
993 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
994 wc->w_cpos, zero_from, zero_to);
996 flush_dcache_page(page);
998 if (ocfs2_should_order_data(inode)) {
999 ret = walk_page_buffers(handle,
1002 ocfs2_journal_dirty_data);
1008 * We don't use generic_commit_write() because we need to
1009 * handle our own i_size update.
1011 ret = block_commit_write(page, from, to);
1016 return copied ? copied : ret;
1020 * Do the actual write of some data into an inode. Optionally allocate
1021 * in order to fulfill the write.
1023 * cpos is the logical cluster offset within the file to write at
1025 * 'phys' is the physical mapping of that offset. a 'phys' value of
1026 * zero indicates that allocation is required. In this case, data_ac
1027 * and meta_ac should be valid (meta_ac can be null if metadata
1028 * allocation isn't required).
1030 static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
1031 struct buffer_head *di_bh,
1032 struct ocfs2_alloc_context *data_ac,
1033 struct ocfs2_alloc_context *meta_ac,
1034 struct ocfs2_write_ctxt *wc)
1036 int ret, i, numpages = 1, new;
1037 unsigned int copied = 0;
1039 u64 v_blkno, p_blkno;
1040 struct address_space *mapping = file->f_mapping;
1041 struct inode *inode = mapping->host;
1042 unsigned long index, start;
1043 struct page **cpages;
1045 new = phys == 0 ? 1 : 0;
1048 * Figure out how many pages we'll be manipulating here. For
1049 * non allocating write, we just change the one
1050 * page. Otherwise, we'll need a whole clusters worth.
1053 numpages = ocfs2_pages_per_cluster(inode->i_sb);
1055 cpages = kzalloc(sizeof(*cpages) * numpages, GFP_NOFS);
1063 * Fill our page array first. That way we've grabbed enough so
1064 * that we can zero and flush if we error after adding the
1068 start = ocfs2_align_clusters_to_page_index(inode->i_sb,
1070 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, wc->w_cpos);
1072 start = wc->w_pos >> PAGE_CACHE_SHIFT;
1073 v_blkno = wc->w_pos >> inode->i_sb->s_blocksize_bits;
1076 for(i = 0; i < numpages; i++) {
1079 cpages[i] = grab_cache_page(mapping, index);
1089 * This is safe to call with the page locks - it won't take
1090 * any additional semaphores or cluster locks.
1092 tmp_pos = wc->w_cpos;
1093 ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
1094 &tmp_pos, 1, di_bh, handle,
1095 data_ac, meta_ac, NULL);
1097 * This shouldn't happen because we must have already
1098 * calculated the correct meta data allocation required. The
1099 * internal tree allocation code should know how to increase
1100 * transaction credits itself.
1102 * If need be, we could handle -EAGAIN for a
1103 * RESTART_TRANS here.
1105 mlog_bug_on_msg(ret == -EAGAIN,
1106 "Inode %llu: EAGAIN return during allocation.\n",
1107 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1114 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL);
1118 * XXX: Should we go readonly here?
1125 BUG_ON(p_blkno == 0);
1127 for(i = 0; i < numpages; i++) {
1128 ret = ocfs2_write_data_page(inode, handle, &p_blkno, cpages[i],
1139 for(i = 0; i < numpages; i++) {
1140 unlock_page(cpages[i]);
1141 mark_page_accessed(cpages[i]);
1142 page_cache_release(cpages[i]);
1146 return copied ? copied : ret;
1149 static void ocfs2_write_ctxt_init(struct ocfs2_write_ctxt *wc,
1150 struct ocfs2_super *osb, loff_t pos,
1151 size_t count, ocfs2_page_writer *cb,
1154 wc->w_count = count;
1156 wc->w_cpos = wc->w_pos >> osb->s_clustersize_bits;
1157 wc->w_finished_copy = 0;
1159 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
1160 wc->w_large_pages = 1;
1162 wc->w_large_pages = 0;
1164 wc->w_write_data_page = cb;
1165 wc->w_private = cb_priv;
1169 * Write a cluster to an inode. The cluster may not be allocated yet,
1170 * in which case it will be. This only exists for buffered writes -
1171 * O_DIRECT takes a more "traditional" path through the kernel.
1173 * The caller is responsible for incrementing pos, written counts, etc
1175 * For file systems that don't support sparse files, pre-allocation
1176 * and page zeroing up until cpos should be done prior to this
1179 * Callers should be holding i_sem, and the rw cluster lock.
1181 * Returns the number of user bytes written, or less than zero for
1184 ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
1185 size_t count, ocfs2_page_writer *actor,
1188 int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
1189 ssize_t written = 0;
1191 struct inode *inode = file->f_mapping->host;
1192 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1193 struct buffer_head *di_bh = NULL;
1194 struct ocfs2_dinode *di;
1195 struct ocfs2_alloc_context *data_ac = NULL;
1196 struct ocfs2_alloc_context *meta_ac = NULL;
1198 struct ocfs2_write_ctxt wc;
1200 ocfs2_write_ctxt_init(&wc, osb, pos, count, actor, priv);
1202 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1207 di = (struct ocfs2_dinode *)di_bh->b_data;
1210 * Take alloc sem here to prevent concurrent lookups. That way
1211 * the mapping, zeroing and tree manipulation within
1212 * ocfs2_write() will be safe against ->readpage(). This
1213 * should also serve to lock out allocation from a shared
1216 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1218 ret = ocfs2_get_clusters(inode, wc.w_cpos, &phys, NULL);
1224 /* phys == 0 means that allocation is required. */
1226 ret = ocfs2_lock_allocators(inode, di, 1, &data_ac, &meta_ac);
1232 credits = ocfs2_calc_extend_credits(inode->i_sb, di, 1);
1235 ret = ocfs2_data_lock(inode, 1);
1241 handle = ocfs2_start_trans(osb, credits);
1242 if (IS_ERR(handle)) {
1243 ret = PTR_ERR(handle);
1248 written = ocfs2_write(file, phys, handle, di_bh, data_ac,
1256 ret = ocfs2_journal_access(handle, inode, di_bh,
1257 OCFS2_JOURNAL_ACCESS_WRITE);
1264 if (pos > inode->i_size) {
1265 i_size_write(inode, pos);
1266 mark_inode_dirty(inode);
1268 inode->i_blocks = ocfs2_align_bytes_to_sectors((u64)(i_size_read(inode)));
1269 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1270 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1271 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1272 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1274 ret = ocfs2_journal_dirty(handle, di_bh);
1279 ocfs2_commit_trans(osb, handle);
1282 ocfs2_data_unlock(inode, 1);
1285 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1286 ocfs2_meta_unlock(inode, 1);
1291 ocfs2_free_alloc_context(data_ac);
1293 ocfs2_free_alloc_context(meta_ac);
1295 return written ? written : ret;
1298 const struct address_space_operations ocfs2_aops = {
1299 .readpage = ocfs2_readpage,
1300 .writepage = ocfs2_writepage,
1302 .sync_page = block_sync_page,
1303 .direct_IO = ocfs2_direct_IO,
1304 .invalidatepage = ocfs2_invalidatepage,
1305 .releasepage = ocfs2_releasepage,
1306 .migratepage = buffer_migrate_page,