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 unsigned int ext_flags;
141 u64 p_blkno, past_eof;
142 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
144 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
145 (unsigned long long)iblock, bh_result, create);
147 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
148 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
149 inode, inode->i_ino);
151 if (S_ISLNK(inode->i_mode)) {
152 /* this always does I/O for some reason. */
153 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
157 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL,
160 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
161 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
162 (unsigned long long)p_blkno);
167 * ocfs2 never allocates in this function - the only time we
168 * need to use BH_New is when we're extending i_size on a file
169 * system which doesn't support holes, in which case BH_New
170 * allows block_prepare_write() to zero.
172 mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb),
173 "ino %lu, iblock %llu\n", inode->i_ino,
174 (unsigned long long)iblock);
176 /* Treat the unwritten extent as a hole for zeroing purposes. */
177 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
178 map_bh(bh_result, inode->i_sb, p_blkno);
180 if (!ocfs2_sparse_alloc(osb)) {
184 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
185 (unsigned long long)iblock,
186 (unsigned long long)p_blkno,
187 (unsigned long long)OCFS2_I(inode)->ip_blkno);
188 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
192 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
193 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
194 (unsigned long long)past_eof);
196 if (create && (iblock >= past_eof))
197 set_buffer_new(bh_result);
208 static int ocfs2_readpage(struct file *file, struct page *page)
210 struct inode *inode = page->mapping->host;
211 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
214 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
216 ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page);
218 if (ret == AOP_TRUNCATED_PAGE)
224 down_read(&OCFS2_I(inode)->ip_alloc_sem);
227 * i_size might have just been updated as we grabed the meta lock. We
228 * might now be discovering a truncate that hit on another node.
229 * block_read_full_page->get_block freaks out if it is asked to read
230 * beyond the end of a file, so we check here. Callers
231 * (generic_file_read, fault->nopage) are clever enough to check i_size
232 * and notice that the page they just read isn't needed.
234 * XXX sys_readahead() seems to get that wrong?
236 if (start >= i_size_read(inode)) {
237 char *addr = kmap(page);
238 memset(addr, 0, PAGE_SIZE);
239 flush_dcache_page(page);
241 SetPageUptodate(page);
246 ret = ocfs2_data_lock_with_page(inode, 0, page);
248 if (ret == AOP_TRUNCATED_PAGE)
254 ret = block_read_full_page(page, ocfs2_get_block);
257 ocfs2_data_unlock(inode, 0);
259 up_read(&OCFS2_I(inode)->ip_alloc_sem);
260 ocfs2_meta_unlock(inode, 0);
268 /* Note: Because we don't support holes, our allocation has
269 * already happened (allocation writes zeros to the file data)
270 * so we don't have to worry about ordered writes in
273 * ->writepage is called during the process of invalidating the page cache
274 * during blocked lock processing. It can't block on any cluster locks
275 * to during block mapping. It's relying on the fact that the block
276 * mapping can't have disappeared under the dirty pages that it is
277 * being asked to write back.
279 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
283 mlog_entry("(0x%p)\n", page);
285 ret = block_write_full_page(page, ocfs2_get_block, wbc);
293 * This is called from ocfs2_write_zero_page() which has handled it's
294 * own cluster locking and has ensured allocation exists for those
295 * blocks to be written.
297 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
298 unsigned from, unsigned to)
302 down_read(&OCFS2_I(inode)->ip_alloc_sem);
304 ret = block_prepare_write(page, from, to, ocfs2_get_block);
306 up_read(&OCFS2_I(inode)->ip_alloc_sem);
311 /* Taken from ext3. We don't necessarily need the full blown
312 * functionality yet, but IMHO it's better to cut and paste the whole
313 * thing so we can avoid introducing our own bugs (and easily pick up
314 * their fixes when they happen) --Mark */
315 int walk_page_buffers( handle_t *handle,
316 struct buffer_head *head,
320 int (*fn)( handle_t *handle,
321 struct buffer_head *bh))
323 struct buffer_head *bh;
324 unsigned block_start, block_end;
325 unsigned blocksize = head->b_size;
327 struct buffer_head *next;
329 for ( bh = head, block_start = 0;
330 ret == 0 && (bh != head || !block_start);
331 block_start = block_end, bh = next)
333 next = bh->b_this_page;
334 block_end = block_start + blocksize;
335 if (block_end <= from || block_start >= to) {
336 if (partial && !buffer_uptodate(bh))
340 err = (*fn)(handle, bh);
347 handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
352 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
353 handle_t *handle = NULL;
356 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
363 if (ocfs2_should_order_data(inode)) {
364 ret = walk_page_buffers(handle,
367 ocfs2_journal_dirty_data);
374 ocfs2_commit_trans(osb, handle);
375 handle = ERR_PTR(ret);
380 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
385 struct inode *inode = mapping->host;
387 mlog_entry("(block = %llu)\n", (unsigned long long)block);
389 /* We don't need to lock journal system files, since they aren't
390 * accessed concurrently from multiple nodes.
392 if (!INODE_JOURNAL(inode)) {
393 err = ocfs2_meta_lock(inode, NULL, 0);
399 down_read(&OCFS2_I(inode)->ip_alloc_sem);
402 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, NULL);
404 if (!INODE_JOURNAL(inode)) {
405 up_read(&OCFS2_I(inode)->ip_alloc_sem);
406 ocfs2_meta_unlock(inode, 0);
410 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
411 (unsigned long long)block);
418 status = err ? 0 : p_blkno;
420 mlog_exit((int)status);
426 * TODO: Make this into a generic get_blocks function.
428 * From do_direct_io in direct-io.c:
429 * "So what we do is to permit the ->get_blocks function to populate
430 * bh.b_size with the size of IO which is permitted at this offset and
433 * This function is called directly from get_more_blocks in direct-io.c.
435 * called like this: dio->get_blocks(dio->inode, fs_startblk,
436 * fs_count, map_bh, dio->rw == WRITE);
438 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
439 struct buffer_head *bh_result, int create)
442 u64 p_blkno, inode_blocks, contig_blocks;
443 unsigned int ext_flags;
444 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
445 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
447 /* This function won't even be called if the request isn't all
448 * nicely aligned and of the right size, so there's no need
449 * for us to check any of that. */
451 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
454 * Any write past EOF is not allowed because we'd be extending.
456 if (create && (iblock + max_blocks) > inode_blocks) {
461 /* This figures out the size of the next contiguous block, and
462 * our logical offset */
463 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
464 &contig_blocks, &ext_flags);
466 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
467 (unsigned long long)iblock);
472 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) {
473 ocfs2_error(inode->i_sb,
474 "Inode %llu has a hole at block %llu\n",
475 (unsigned long long)OCFS2_I(inode)->ip_blkno,
476 (unsigned long long)iblock);
482 * get_more_blocks() expects us to describe a hole by clearing
483 * the mapped bit on bh_result().
485 * Consider an unwritten extent as a hole.
487 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
488 map_bh(bh_result, inode->i_sb, p_blkno);
491 * ocfs2_prepare_inode_for_write() should have caught
492 * the case where we'd be filling a hole and triggered
493 * a buffered write instead.
501 clear_buffer_mapped(bh_result);
504 /* make sure we don't map more than max_blocks blocks here as
505 that's all the kernel will handle at this point. */
506 if (max_blocks < contig_blocks)
507 contig_blocks = max_blocks;
508 bh_result->b_size = contig_blocks << blocksize_bits;
514 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
515 * particularly interested in the aio/dio case. Like the core uses
516 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
517 * truncation on another.
519 static void ocfs2_dio_end_io(struct kiocb *iocb,
524 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
526 /* this io's submitter should not have unlocked this before we could */
527 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
528 ocfs2_iocb_clear_rw_locked(iocb);
529 up_read(&inode->i_alloc_sem);
530 ocfs2_rw_unlock(inode, 0);
534 * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
535 * from ext3. PageChecked() bits have been removed as OCFS2 does not
536 * do journalled data.
538 static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
540 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
542 journal_invalidatepage(journal, page, offset);
545 static int ocfs2_releasepage(struct page *page, gfp_t wait)
547 journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal;
549 if (!page_has_buffers(page))
551 return journal_try_to_free_buffers(journal, page, wait);
554 static ssize_t ocfs2_direct_IO(int rw,
556 const struct iovec *iov,
558 unsigned long nr_segs)
560 struct file *file = iocb->ki_filp;
561 struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host;
566 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
568 * We get PR data locks even for O_DIRECT. This
569 * allows concurrent O_DIRECT I/O but doesn't let
570 * O_DIRECT with extending and buffered zeroing writes
571 * race. If they did race then the buffered zeroing
572 * could be written back after the O_DIRECT I/O. It's
573 * one thing to tell people not to mix buffered and
574 * O_DIRECT writes, but expecting them to understand
575 * that file extension is also an implicit buffered
576 * write is too much. By getting the PR we force
577 * writeback of the buffered zeroing before
580 ret = ocfs2_data_lock(inode, 0);
585 ocfs2_data_unlock(inode, 0);
588 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
589 inode->i_sb->s_bdev, iov, offset,
591 ocfs2_direct_IO_get_blocks,
598 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
603 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
605 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
608 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
610 cluster_start = cpos % cpp;
611 cluster_start = cluster_start << osb->s_clustersize_bits;
613 cluster_end = cluster_start + osb->s_clustersize;
616 BUG_ON(cluster_start > PAGE_SIZE);
617 BUG_ON(cluster_end > PAGE_SIZE);
620 *start = cluster_start;
626 * 'from' and 'to' are the region in the page to avoid zeroing.
628 * If pagesize > clustersize, this function will avoid zeroing outside
629 * of the cluster boundary.
631 * from == to == 0 is code for "zero the entire cluster region"
633 static void ocfs2_clear_page_regions(struct page *page,
634 struct ocfs2_super *osb, u32 cpos,
635 unsigned from, unsigned to)
638 unsigned int cluster_start, cluster_end;
640 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
642 kaddr = kmap_atomic(page, KM_USER0);
645 if (from > cluster_start)
646 memset(kaddr + cluster_start, 0, from - cluster_start);
647 if (to < cluster_end)
648 memset(kaddr + to, 0, cluster_end - to);
650 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
653 kunmap_atomic(kaddr, KM_USER0);
657 * Some of this taken from block_prepare_write(). We already have our
658 * mapping by now though, and the entire write will be allocating or
659 * it won't, so not much need to use BH_New.
661 * This will also skip zeroing, which is handled externally.
663 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
664 struct inode *inode, unsigned int from,
665 unsigned int to, int new)
668 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
669 unsigned int block_end, block_start;
670 unsigned int bsize = 1 << inode->i_blkbits;
672 if (!page_has_buffers(page))
673 create_empty_buffers(page, bsize, 0);
675 head = page_buffers(page);
676 for (bh = head, block_start = 0; bh != head || !block_start;
677 bh = bh->b_this_page, block_start += bsize) {
678 block_end = block_start + bsize;
681 * Ignore blocks outside of our i/o range -
682 * they may belong to unallocated clusters.
684 if (block_start >= to || block_end <= from) {
685 if (PageUptodate(page))
686 set_buffer_uptodate(bh);
691 * For an allocating write with cluster size >= page
692 * size, we always write the entire page.
696 clear_buffer_new(bh);
698 if (!buffer_mapped(bh)) {
699 map_bh(bh, inode->i_sb, *p_blkno);
700 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
703 if (PageUptodate(page)) {
704 if (!buffer_uptodate(bh))
705 set_buffer_uptodate(bh);
706 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
707 (block_start < from || block_end > to)) {
708 ll_rw_block(READ, 1, &bh);
712 *p_blkno = *p_blkno + 1;
716 * If we issued read requests - let them complete.
718 while(wait_bh > wait) {
719 wait_on_buffer(*--wait_bh);
720 if (!buffer_uptodate(*wait_bh))
724 if (ret == 0 || !new)
728 * If we get -EIO above, zero out any newly allocated blocks
729 * to avoid exposing stale data.
736 block_end = block_start + bsize;
737 if (block_end <= from)
739 if (block_start >= to)
742 kaddr = kmap_atomic(page, KM_USER0);
743 memset(kaddr+block_start, 0, bh->b_size);
744 flush_dcache_page(page);
745 kunmap_atomic(kaddr, KM_USER0);
746 set_buffer_uptodate(bh);
747 mark_buffer_dirty(bh);
750 block_start = block_end;
751 bh = bh->b_this_page;
752 } while (bh != head);
758 * This will copy user data from the buffer page in the splice
761 * For now, we ignore SPLICE_F_MOVE as that would require some extra
762 * communication out all the way to ocfs2_write().
764 int ocfs2_map_and_write_splice_data(struct inode *inode,
765 struct ocfs2_write_ctxt *wc, u64 *p_blkno,
766 unsigned int *ret_from, unsigned int *ret_to)
769 unsigned int to, from, cluster_start, cluster_end;
771 struct ocfs2_splice_write_priv *sp = wc->w_private;
772 struct pipe_buffer *buf = sp->s_buf;
773 unsigned long bytes, src_from;
774 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
776 ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
780 src_from = sp->s_buf_offset;
783 if (wc->w_large_pages) {
785 * For cluster size < page size, we have to
786 * calculate pos within the cluster and obey
787 * the rightmost boundary.
789 bytes = min(bytes, (unsigned long)(osb->s_clustersize
790 - (wc->w_pos & (osb->s_clustersize - 1))));
794 if (wc->w_this_page_new)
795 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
796 cluster_start, cluster_end, 1);
798 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
805 BUG_ON(from > PAGE_CACHE_SIZE);
806 BUG_ON(to > PAGE_CACHE_SIZE);
807 BUG_ON(from > osb->s_clustersize);
808 BUG_ON(to > osb->s_clustersize);
810 src = buf->ops->map(sp->s_pipe, buf, 1);
811 dst = kmap_atomic(wc->w_this_page, KM_USER1);
812 memcpy(dst + from, src + src_from, bytes);
813 kunmap_atomic(wc->w_this_page, KM_USER1);
814 buf->ops->unmap(sp->s_pipe, buf, src);
816 wc->w_finished_copy = 1;
822 return bytes ? (unsigned int)bytes : ret;
826 * This will copy user data from the iovec in the buffered write
829 int ocfs2_map_and_write_user_data(struct inode *inode,
830 struct ocfs2_write_ctxt *wc, u64 *p_blkno,
831 unsigned int *ret_from, unsigned int *ret_to)
834 unsigned int to, from, cluster_start, cluster_end;
835 unsigned long bytes, src_from;
837 struct ocfs2_buffered_write_priv *bp = wc->w_private;
838 const struct iovec *cur_iov = bp->b_cur_iov;
840 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
842 ocfs2_figure_cluster_boundaries(osb, wc->w_cpos, &cluster_start,
845 buf = cur_iov->iov_base + bp->b_cur_off;
846 src_from = (unsigned long)buf & ~PAGE_CACHE_MASK;
848 from = wc->w_pos & (PAGE_CACHE_SIZE - 1);
851 * This is a lot of comparisons, but it reads quite
852 * easily, which is important here.
854 /* Stay within the src page */
855 bytes = PAGE_SIZE - src_from;
856 /* Stay within the vector */
858 (unsigned long)(cur_iov->iov_len - bp->b_cur_off));
859 /* Stay within count */
860 bytes = min(bytes, (unsigned long)wc->w_count);
862 * For clustersize > page size, just stay within
863 * target page, otherwise we have to calculate pos
864 * within the cluster and obey the rightmost
867 if (wc->w_large_pages) {
869 * For cluster size < page size, we have to
870 * calculate pos within the cluster and obey
871 * the rightmost boundary.
873 bytes = min(bytes, (unsigned long)(osb->s_clustersize
874 - (wc->w_pos & (osb->s_clustersize - 1))));
877 * cluster size > page size is the most common
878 * case - we just stay within the target page
881 bytes = min(bytes, PAGE_CACHE_SIZE - from);
886 if (wc->w_this_page_new)
887 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
888 cluster_start, cluster_end, 1);
890 ret = ocfs2_map_page_blocks(wc->w_this_page, p_blkno, inode,
897 BUG_ON(from > PAGE_CACHE_SIZE);
898 BUG_ON(to > PAGE_CACHE_SIZE);
899 BUG_ON(from > osb->s_clustersize);
900 BUG_ON(to > osb->s_clustersize);
902 dst = kmap(wc->w_this_page);
903 memcpy(dst + from, bp->b_src_buf + src_from, bytes);
904 kunmap(wc->w_this_page);
907 * XXX: This is slow, but simple. The caller of
908 * ocfs2_buffered_write_cluster() is responsible for
909 * passing through the iovecs, so it's difficult to
910 * predict what our next step is in here after our
911 * initial write. A future version should be pushing
912 * that iovec manipulation further down.
914 * By setting this, we indicate that a copy from user
915 * data was done, and subsequent calls for this
916 * cluster will skip copying more data.
918 wc->w_finished_copy = 1;
924 return bytes ? (unsigned int)bytes : ret;
928 * Map, fill and write a page to disk.
930 * The work of copying data is done via callback. Newly allocated
931 * pages which don't take user data will be zero'd (set 'new' to
932 * indicate an allocating write)
934 * Returns a negative error code or the number of bytes copied into
937 int ocfs2_write_data_page(struct inode *inode, handle_t *handle,
938 u64 *p_blkno, struct page *page,
939 struct ocfs2_write_ctxt *wc, int new)
942 unsigned int from = 0, to = 0;
943 unsigned int cluster_start, cluster_end;
944 unsigned int zero_from = 0, zero_to = 0;
946 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), wc->w_cpos,
947 &cluster_start, &cluster_end);
949 if ((wc->w_pos >> PAGE_CACHE_SHIFT) == page->index
950 && !wc->w_finished_copy) {
952 wc->w_this_page = page;
953 wc->w_this_page_new = new;
954 ret = wc->w_write_data_page(inode, wc, p_blkno, &from, &to);
965 from = cluster_start;
970 * If we haven't allocated the new page yet, we
971 * shouldn't be writing it out without copying user
972 * data. This is likely a math error from the caller.
976 from = cluster_start;
979 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
980 cluster_start, cluster_end, 1);
988 * Parts of newly allocated pages need to be zero'd.
990 * Above, we have also rewritten 'to' and 'from' - as far as
991 * the rest of the function is concerned, the entire cluster
992 * range inside of a page needs to be written.
994 * We can skip this if the page is up to date - it's already
995 * been zero'd from being read in as a hole.
997 if (new && !PageUptodate(page))
998 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
999 wc->w_cpos, zero_from, zero_to);
1001 flush_dcache_page(page);
1003 if (ocfs2_should_order_data(inode)) {
1004 ret = walk_page_buffers(handle,
1007 ocfs2_journal_dirty_data);
1013 * We don't use generic_commit_write() because we need to
1014 * handle our own i_size update.
1016 ret = block_commit_write(page, from, to);
1021 return copied ? copied : ret;
1025 * Do the actual write of some data into an inode. Optionally allocate
1026 * in order to fulfill the write.
1028 * cpos is the logical cluster offset within the file to write at
1030 * 'phys' is the physical mapping of that offset. a 'phys' value of
1031 * zero indicates that allocation is required. In this case, data_ac
1032 * and meta_ac should be valid (meta_ac can be null if metadata
1033 * allocation isn't required).
1035 static ssize_t ocfs2_write(struct file *file, u32 phys, handle_t *handle,
1036 struct buffer_head *di_bh,
1037 struct ocfs2_alloc_context *data_ac,
1038 struct ocfs2_alloc_context *meta_ac,
1039 struct ocfs2_write_ctxt *wc)
1041 int ret, i, numpages = 1, new;
1042 unsigned int copied = 0;
1044 u64 v_blkno, p_blkno;
1045 struct address_space *mapping = file->f_mapping;
1046 struct inode *inode = mapping->host;
1047 unsigned long index, start;
1048 struct page **cpages;
1050 new = phys == 0 ? 1 : 0;
1053 * Figure out how many pages we'll be manipulating here. For
1054 * non allocating write, we just change the one
1055 * page. Otherwise, we'll need a whole clusters worth.
1058 numpages = ocfs2_pages_per_cluster(inode->i_sb);
1060 cpages = kzalloc(sizeof(*cpages) * numpages, GFP_NOFS);
1068 * Fill our page array first. That way we've grabbed enough so
1069 * that we can zero and flush if we error after adding the
1073 start = ocfs2_align_clusters_to_page_index(inode->i_sb,
1075 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, wc->w_cpos);
1077 start = wc->w_pos >> PAGE_CACHE_SHIFT;
1078 v_blkno = wc->w_pos >> inode->i_sb->s_blocksize_bits;
1081 for(i = 0; i < numpages; i++) {
1084 cpages[i] = grab_cache_page(mapping, index);
1094 * This is safe to call with the page locks - it won't take
1095 * any additional semaphores or cluster locks.
1097 tmp_pos = wc->w_cpos;
1098 ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
1099 &tmp_pos, 1, di_bh, handle,
1100 data_ac, meta_ac, NULL);
1102 * This shouldn't happen because we must have already
1103 * calculated the correct meta data allocation required. The
1104 * internal tree allocation code should know how to increase
1105 * transaction credits itself.
1107 * If need be, we could handle -EAGAIN for a
1108 * RESTART_TRANS here.
1110 mlog_bug_on_msg(ret == -EAGAIN,
1111 "Inode %llu: EAGAIN return during allocation.\n",
1112 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1119 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1124 * XXX: Should we go readonly here?
1131 BUG_ON(p_blkno == 0);
1133 for(i = 0; i < numpages; i++) {
1134 ret = ocfs2_write_data_page(inode, handle, &p_blkno, cpages[i],
1145 for(i = 0; i < numpages; i++) {
1146 unlock_page(cpages[i]);
1147 mark_page_accessed(cpages[i]);
1148 page_cache_release(cpages[i]);
1152 return copied ? copied : ret;
1155 static void ocfs2_write_ctxt_init(struct ocfs2_write_ctxt *wc,
1156 struct ocfs2_super *osb, loff_t pos,
1157 size_t count, ocfs2_page_writer *cb,
1160 wc->w_count = count;
1162 wc->w_cpos = wc->w_pos >> osb->s_clustersize_bits;
1163 wc->w_finished_copy = 0;
1165 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
1166 wc->w_large_pages = 1;
1168 wc->w_large_pages = 0;
1170 wc->w_write_data_page = cb;
1171 wc->w_private = cb_priv;
1175 * Write a cluster to an inode. The cluster may not be allocated yet,
1176 * in which case it will be. This only exists for buffered writes -
1177 * O_DIRECT takes a more "traditional" path through the kernel.
1179 * The caller is responsible for incrementing pos, written counts, etc
1181 * For file systems that don't support sparse files, pre-allocation
1182 * and page zeroing up until cpos should be done prior to this
1185 * Callers should be holding i_sem, and the rw cluster lock.
1187 * Returns the number of user bytes written, or less than zero for
1190 ssize_t ocfs2_buffered_write_cluster(struct file *file, loff_t pos,
1191 size_t count, ocfs2_page_writer *actor,
1194 int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
1195 ssize_t written = 0;
1197 struct inode *inode = file->f_mapping->host;
1198 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1199 struct buffer_head *di_bh = NULL;
1200 struct ocfs2_dinode *di;
1201 struct ocfs2_alloc_context *data_ac = NULL;
1202 struct ocfs2_alloc_context *meta_ac = NULL;
1204 struct ocfs2_write_ctxt wc;
1206 ocfs2_write_ctxt_init(&wc, osb, pos, count, actor, priv);
1208 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1213 di = (struct ocfs2_dinode *)di_bh->b_data;
1216 * Take alloc sem here to prevent concurrent lookups. That way
1217 * the mapping, zeroing and tree manipulation within
1218 * ocfs2_write() will be safe against ->readpage(). This
1219 * should also serve to lock out allocation from a shared
1222 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1224 ret = ocfs2_get_clusters(inode, wc.w_cpos, &phys, NULL, NULL);
1230 /* phys == 0 means that allocation is required. */
1232 ret = ocfs2_lock_allocators(inode, di, 1, &data_ac, &meta_ac);
1238 credits = ocfs2_calc_extend_credits(inode->i_sb, di, 1);
1241 ret = ocfs2_data_lock(inode, 1);
1247 handle = ocfs2_start_trans(osb, credits);
1248 if (IS_ERR(handle)) {
1249 ret = PTR_ERR(handle);
1254 written = ocfs2_write(file, phys, handle, di_bh, data_ac,
1262 ret = ocfs2_journal_access(handle, inode, di_bh,
1263 OCFS2_JOURNAL_ACCESS_WRITE);
1270 if (pos > inode->i_size) {
1271 i_size_write(inode, pos);
1272 mark_inode_dirty(inode);
1274 inode->i_blocks = ocfs2_inode_sector_count(inode);
1275 di->i_size = cpu_to_le64((u64)i_size_read(inode));
1276 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1277 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
1278 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1280 ret = ocfs2_journal_dirty(handle, di_bh);
1285 ocfs2_commit_trans(osb, handle);
1288 ocfs2_data_unlock(inode, 1);
1291 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1292 ocfs2_meta_unlock(inode, 1);
1297 ocfs2_free_alloc_context(data_ac);
1299 ocfs2_free_alloc_context(meta_ac);
1301 return written ? written : ret;
1304 const struct address_space_operations ocfs2_aops = {
1305 .readpage = ocfs2_readpage,
1306 .writepage = ocfs2_writepage,
1308 .sync_page = block_sync_page,
1309 .direct_IO = ocfs2_direct_IO,
1310 .invalidatepage = ocfs2_invalidatepage,
1311 .releasepage = ocfs2_releasepage,
1312 .migratepage = buffer_migrate_page,