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>
31 #include <linux/blkdev.h>
33 #include <cluster/masklog.h>
40 #include "extent_map.h"
47 #include "refcounttree.h"
48 #include "ocfs2_trace.h"
50 #include "buffer_head_io.h"
55 static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
56 struct buffer_head *bh_result, int create)
60 struct ocfs2_dinode *fe = NULL;
61 struct buffer_head *bh = NULL;
62 struct buffer_head *buffer_cache_bh = NULL;
63 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
66 trace_ocfs2_symlink_get_block(
67 (unsigned long long)OCFS2_I(inode)->ip_blkno,
68 (unsigned long long)iblock, bh_result, create);
70 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
72 if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
73 mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
74 (unsigned long long)iblock);
78 status = ocfs2_read_inode_block(inode, &bh);
83 fe = (struct ocfs2_dinode *) bh->b_data;
85 if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
86 le32_to_cpu(fe->i_clusters))) {
88 mlog(ML_ERROR, "block offset is outside the allocated size: "
89 "%llu\n", (unsigned long long)iblock);
93 /* We don't use the page cache to create symlink data, so if
94 * need be, copy it over from the buffer cache. */
95 if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
96 u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
98 buffer_cache_bh = sb_getblk(osb->sb, blkno);
99 if (!buffer_cache_bh) {
101 mlog(ML_ERROR, "couldn't getblock for symlink!\n");
105 /* we haven't locked out transactions, so a commit
106 * could've happened. Since we've got a reference on
107 * the bh, even if it commits while we're doing the
108 * copy, the data is still good. */
109 if (buffer_jbd(buffer_cache_bh)
110 && ocfs2_inode_is_new(inode)) {
111 kaddr = kmap_atomic(bh_result->b_page);
113 mlog(ML_ERROR, "couldn't kmap!\n");
116 memcpy(kaddr + (bh_result->b_size * iblock),
117 buffer_cache_bh->b_data,
119 kunmap_atomic(kaddr);
120 set_buffer_uptodate(bh_result);
122 brelse(buffer_cache_bh);
125 map_bh(bh_result, inode->i_sb,
126 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
136 int ocfs2_get_block(struct inode *inode, sector_t iblock,
137 struct buffer_head *bh_result, int create)
140 unsigned int ext_flags;
141 u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
142 u64 p_blkno, count, past_eof;
143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
145 trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
146 (unsigned long long)iblock, bh_result, create);
148 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
149 mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
150 inode, inode->i_ino);
152 if (S_ISLNK(inode->i_mode)) {
153 /* this always does I/O for some reason. */
154 err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
158 err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
161 mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
162 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
163 (unsigned long long)p_blkno);
167 if (max_blocks < count)
171 * ocfs2 never allocates in this function - the only time we
172 * need to use BH_New is when we're extending i_size on a file
173 * system which doesn't support holes, in which case BH_New
174 * allows __block_write_begin() to zero.
176 * If we see this on a sparse file system, then a truncate has
177 * raced us and removed the cluster. In this case, we clear
178 * the buffers dirty and uptodate bits and let the buffer code
179 * ignore it as a hole.
181 if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
182 clear_buffer_dirty(bh_result);
183 clear_buffer_uptodate(bh_result);
187 /* Treat the unwritten extent as a hole for zeroing purposes. */
188 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
189 map_bh(bh_result, inode->i_sb, p_blkno);
191 bh_result->b_size = count << inode->i_blkbits;
193 if (!ocfs2_sparse_alloc(osb)) {
197 "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
198 (unsigned long long)iblock,
199 (unsigned long long)p_blkno,
200 (unsigned long long)OCFS2_I(inode)->ip_blkno);
201 mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
207 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
209 trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
210 (unsigned long long)past_eof);
211 if (create && (iblock >= past_eof))
212 set_buffer_new(bh_result);
221 int ocfs2_read_inline_data(struct inode *inode, struct page *page,
222 struct buffer_head *di_bh)
226 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
228 if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
229 ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
234 size = i_size_read(inode);
236 if (size > PAGE_CACHE_SIZE ||
237 size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
238 ocfs2_error(inode->i_sb,
239 "Inode %llu has with inline data has bad size: %Lu",
240 (unsigned long long)OCFS2_I(inode)->ip_blkno,
241 (unsigned long long)size);
245 kaddr = kmap_atomic(page);
247 memcpy(kaddr, di->id2.i_data.id_data, size);
248 /* Clear the remaining part of the page */
249 memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
250 flush_dcache_page(page);
251 kunmap_atomic(kaddr);
253 SetPageUptodate(page);
258 static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
261 struct buffer_head *di_bh = NULL;
263 BUG_ON(!PageLocked(page));
264 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
266 ret = ocfs2_read_inode_block(inode, &di_bh);
272 ret = ocfs2_read_inline_data(inode, page, di_bh);
280 static int ocfs2_readpage(struct file *file, struct page *page)
282 struct inode *inode = page->mapping->host;
283 struct ocfs2_inode_info *oi = OCFS2_I(inode);
284 loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
287 trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
288 (page ? page->index : 0));
290 ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
292 if (ret == AOP_TRUNCATED_PAGE)
298 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
300 * Unlock the page and cycle ip_alloc_sem so that we don't
301 * busyloop waiting for ip_alloc_sem to unlock
303 ret = AOP_TRUNCATED_PAGE;
306 down_read(&oi->ip_alloc_sem);
307 up_read(&oi->ip_alloc_sem);
308 goto out_inode_unlock;
312 * i_size might have just been updated as we grabed the meta lock. We
313 * might now be discovering a truncate that hit on another node.
314 * block_read_full_page->get_block freaks out if it is asked to read
315 * beyond the end of a file, so we check here. Callers
316 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
317 * and notice that the page they just read isn't needed.
319 * XXX sys_readahead() seems to get that wrong?
321 if (start >= i_size_read(inode)) {
322 zero_user(page, 0, PAGE_SIZE);
323 SetPageUptodate(page);
328 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
329 ret = ocfs2_readpage_inline(inode, page);
331 ret = block_read_full_page(page, ocfs2_get_block);
335 up_read(&OCFS2_I(inode)->ip_alloc_sem);
337 ocfs2_inode_unlock(inode, 0);
345 * This is used only for read-ahead. Failures or difficult to handle
346 * situations are safe to ignore.
348 * Right now, we don't bother with BH_Boundary - in-inode extent lists
349 * are quite large (243 extents on 4k blocks), so most inodes don't
350 * grow out to a tree. If need be, detecting boundary extents could
351 * trivially be added in a future version of ocfs2_get_block().
353 static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
354 struct list_head *pages, unsigned nr_pages)
357 struct inode *inode = mapping->host;
358 struct ocfs2_inode_info *oi = OCFS2_I(inode);
363 * Use the nonblocking flag for the dlm code to avoid page
364 * lock inversion, but don't bother with retrying.
366 ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
370 if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
371 ocfs2_inode_unlock(inode, 0);
376 * Don't bother with inline-data. There isn't anything
377 * to read-ahead in that case anyway...
379 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
383 * Check whether a remote node truncated this file - we just
384 * drop out in that case as it's not worth handling here.
386 last = list_entry(pages->prev, struct page, lru);
387 start = (loff_t)last->index << PAGE_CACHE_SHIFT;
388 if (start >= i_size_read(inode))
391 err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
394 up_read(&oi->ip_alloc_sem);
395 ocfs2_inode_unlock(inode, 0);
400 /* Note: Because we don't support holes, our allocation has
401 * already happened (allocation writes zeros to the file data)
402 * so we don't have to worry about ordered writes in
405 * ->writepage is called during the process of invalidating the page cache
406 * during blocked lock processing. It can't block on any cluster locks
407 * to during block mapping. It's relying on the fact that the block
408 * mapping can't have disappeared under the dirty pages that it is
409 * being asked to write back.
411 static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
413 trace_ocfs2_writepage(
414 (unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
417 return block_write_full_page(page, ocfs2_get_block, wbc);
420 /* Taken from ext3. We don't necessarily need the full blown
421 * functionality yet, but IMHO it's better to cut and paste the whole
422 * thing so we can avoid introducing our own bugs (and easily pick up
423 * their fixes when they happen) --Mark */
424 int walk_page_buffers( handle_t *handle,
425 struct buffer_head *head,
429 int (*fn)( handle_t *handle,
430 struct buffer_head *bh))
432 struct buffer_head *bh;
433 unsigned block_start, block_end;
434 unsigned blocksize = head->b_size;
436 struct buffer_head *next;
438 for ( bh = head, block_start = 0;
439 ret == 0 && (bh != head || !block_start);
440 block_start = block_end, bh = next)
442 next = bh->b_this_page;
443 block_end = block_start + blocksize;
444 if (block_end <= from || block_start >= to) {
445 if (partial && !buffer_uptodate(bh))
449 err = (*fn)(handle, bh);
456 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
461 struct inode *inode = mapping->host;
463 trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
464 (unsigned long long)block);
466 /* We don't need to lock journal system files, since they aren't
467 * accessed concurrently from multiple nodes.
469 if (!INODE_JOURNAL(inode)) {
470 err = ocfs2_inode_lock(inode, NULL, 0);
476 down_read(&OCFS2_I(inode)->ip_alloc_sem);
479 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
480 err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
483 if (!INODE_JOURNAL(inode)) {
484 up_read(&OCFS2_I(inode)->ip_alloc_sem);
485 ocfs2_inode_unlock(inode, 0);
489 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
490 (unsigned long long)block);
496 status = err ? 0 : p_blkno;
502 * TODO: Make this into a generic get_blocks function.
504 * From do_direct_io in direct-io.c:
505 * "So what we do is to permit the ->get_blocks function to populate
506 * bh.b_size with the size of IO which is permitted at this offset and
509 * This function is called directly from get_more_blocks in direct-io.c.
511 * called like this: dio->get_blocks(dio->inode, fs_startblk,
512 * fs_count, map_bh, dio->rw == WRITE);
514 * Note that we never bother to allocate blocks here, and thus ignore the
517 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
518 struct buffer_head *bh_result, int create)
521 u64 p_blkno, inode_blocks, contig_blocks;
522 unsigned int ext_flags;
523 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
524 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
526 /* This function won't even be called if the request isn't all
527 * nicely aligned and of the right size, so there's no need
528 * for us to check any of that. */
530 inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
532 /* This figures out the size of the next contiguous block, and
533 * our logical offset */
534 ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
535 &contig_blocks, &ext_flags);
537 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
538 (unsigned long long)iblock);
543 /* We should already CoW the refcounted extent in case of create. */
544 BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
547 * get_more_blocks() expects us to describe a hole by clearing
548 * the mapped bit on bh_result().
550 * Consider an unwritten extent as a hole.
552 if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
553 map_bh(bh_result, inode->i_sb, p_blkno);
555 clear_buffer_mapped(bh_result);
557 /* make sure we don't map more than max_blocks blocks here as
558 that's all the kernel will handle at this point. */
559 if (max_blocks < contig_blocks)
560 contig_blocks = max_blocks;
561 bh_result->b_size = contig_blocks << blocksize_bits;
567 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
568 * particularly interested in the aio/dio case. We use the rw_lock DLM lock
569 * to protect io on one node from truncation on another.
571 static void ocfs2_dio_end_io(struct kiocb *iocb,
576 struct inode *inode = file_inode(iocb->ki_filp);
579 /* this io's submitter should not have unlocked this before we could */
580 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
582 if (ocfs2_iocb_is_sem_locked(iocb))
583 ocfs2_iocb_clear_sem_locked(iocb);
585 if (ocfs2_iocb_is_unaligned_aio(iocb)) {
586 ocfs2_iocb_clear_unaligned_aio(iocb);
588 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
591 ocfs2_iocb_clear_rw_locked(iocb);
593 level = ocfs2_iocb_rw_locked_level(iocb);
594 ocfs2_rw_unlock(inode, level);
597 static int ocfs2_releasepage(struct page *page, gfp_t wait)
599 if (!page_has_buffers(page))
601 return try_to_free_buffers(page);
604 static int ocfs2_is_overwrite(struct ocfs2_super *osb,
605 struct inode *inode, loff_t offset)
610 unsigned int num_clusters = 0;
611 unsigned int ext_flags = 0;
613 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
614 ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
615 &num_clusters, &ext_flags);
621 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN))
627 static ssize_t ocfs2_direct_IO_write(struct kiocb *iocb,
628 struct iov_iter *iter,
633 bool orphaned = false;
634 int is_overwrite = 0;
635 struct file *file = iocb->ki_filp;
636 struct inode *inode = file_inode(file)->i_mapping->host;
637 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
638 struct buffer_head *di_bh = NULL;
639 size_t count = iter->count;
640 journal_t *journal = osb->journal->j_journal;
643 loff_t final_size = offset + count;
644 int append_write = offset >= i_size_read(inode) ? 1 : 0;
645 unsigned int num_clusters = 0;
646 unsigned int ext_flags = 0;
651 zero_len = do_div(o, 1 << osb->s_clustersize_bits);
652 cluster_align = !zero_len;
656 * when final_size > inode->i_size, inode->i_size will be
657 * updated after direct write, so add the inode to orphan
660 if (final_size > i_size_read(inode)) {
661 ret = ocfs2_add_inode_to_orphan(osb, inode);
670 ret = ocfs2_inode_lock(inode, &di_bh, 1);
676 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
677 ret = ocfs2_zero_extend(inode, di_bh, offset);
679 ret = ocfs2_extend_no_holes(inode, di_bh, offset,
683 ocfs2_inode_unlock(inode, 1);
688 is_overwrite = ocfs2_is_overwrite(osb, inode, offset);
689 if (is_overwrite < 0) {
690 mlog_errno(is_overwrite);
691 ocfs2_inode_unlock(inode, 1);
696 ocfs2_inode_unlock(inode, 1);
701 written = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev,
703 ocfs2_direct_IO_get_blocks,
704 ocfs2_dio_end_io, NULL, 0);
705 if (unlikely(written < 0)) {
706 loff_t i_size = i_size_read(inode);
708 if (offset + count > i_size) {
709 ret = ocfs2_inode_lock(inode, &di_bh, 1);
715 if (i_size == i_size_read(inode)) {
716 ret = ocfs2_truncate_file(inode, di_bh,
722 ocfs2_inode_unlock(inode, 1);
728 ocfs2_inode_unlock(inode, 1);
731 ret = jbd2_journal_force_commit(journal);
735 } else if (written < 0 && append_write && !is_overwrite &&
738 u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
740 ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
741 &num_clusters, &ext_flags);
747 BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN));
749 ret = blkdev_issue_zeroout(osb->sb->s_bdev,
750 p_cpos << (osb->s_clustersize_bits - 9),
751 zero_len >> 9, GFP_KERNEL, false);
759 int update_isize = written > 0 ? 1 : 0;
760 loff_t end = update_isize ? offset + written : 0;
762 tmp_ret = ocfs2_del_inode_from_orphan(osb, inode,
769 tmp_ret = jbd2_journal_force_commit(journal);
782 static ssize_t ocfs2_direct_IO(int rw,
784 struct iov_iter *iter,
787 struct file *file = iocb->ki_filp;
788 struct inode *inode = file_inode(file)->i_mapping->host;
789 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
790 int full_coherency = !(osb->s_mount_opt &
791 OCFS2_MOUNT_COHERENCY_BUFFERED);
794 * Fallback to buffered I/O if we see an inode without
797 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
800 /* Fallback to buffered I/O if we are appending and
801 * concurrent O_DIRECT writes are allowed.
803 if (i_size_read(inode) <= offset && !full_coherency)
807 return __blockdev_direct_IO(rw, iocb, inode,
810 ocfs2_direct_IO_get_blocks,
811 ocfs2_dio_end_io, NULL, 0);
813 return ocfs2_direct_IO_write(iocb, iter, offset);
816 static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
821 unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
823 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
826 cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
828 cluster_start = cpos % cpp;
829 cluster_start = cluster_start << osb->s_clustersize_bits;
831 cluster_end = cluster_start + osb->s_clustersize;
834 BUG_ON(cluster_start > PAGE_SIZE);
835 BUG_ON(cluster_end > PAGE_SIZE);
838 *start = cluster_start;
844 * 'from' and 'to' are the region in the page to avoid zeroing.
846 * If pagesize > clustersize, this function will avoid zeroing outside
847 * of the cluster boundary.
849 * from == to == 0 is code for "zero the entire cluster region"
851 static void ocfs2_clear_page_regions(struct page *page,
852 struct ocfs2_super *osb, u32 cpos,
853 unsigned from, unsigned to)
856 unsigned int cluster_start, cluster_end;
858 ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
860 kaddr = kmap_atomic(page);
863 if (from > cluster_start)
864 memset(kaddr + cluster_start, 0, from - cluster_start);
865 if (to < cluster_end)
866 memset(kaddr + to, 0, cluster_end - to);
868 memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
871 kunmap_atomic(kaddr);
875 * Nonsparse file systems fully allocate before we get to the write
876 * code. This prevents ocfs2_write() from tagging the write as an
877 * allocating one, which means ocfs2_map_page_blocks() might try to
878 * read-in the blocks at the tail of our file. Avoid reading them by
879 * testing i_size against each block offset.
881 static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
882 unsigned int block_start)
884 u64 offset = page_offset(page) + block_start;
886 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
889 if (i_size_read(inode) > offset)
896 * Some of this taken from __block_write_begin(). We already have our
897 * mapping by now though, and the entire write will be allocating or
898 * it won't, so not much need to use BH_New.
900 * This will also skip zeroing, which is handled externally.
902 int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
903 struct inode *inode, unsigned int from,
904 unsigned int to, int new)
907 struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
908 unsigned int block_end, block_start;
909 unsigned int bsize = 1 << inode->i_blkbits;
911 if (!page_has_buffers(page))
912 create_empty_buffers(page, bsize, 0);
914 head = page_buffers(page);
915 for (bh = head, block_start = 0; bh != head || !block_start;
916 bh = bh->b_this_page, block_start += bsize) {
917 block_end = block_start + bsize;
919 clear_buffer_new(bh);
922 * Ignore blocks outside of our i/o range -
923 * they may belong to unallocated clusters.
925 if (block_start >= to || block_end <= from) {
926 if (PageUptodate(page))
927 set_buffer_uptodate(bh);
932 * For an allocating write with cluster size >= page
933 * size, we always write the entire page.
938 if (!buffer_mapped(bh)) {
939 map_bh(bh, inode->i_sb, *p_blkno);
940 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
943 if (PageUptodate(page)) {
944 if (!buffer_uptodate(bh))
945 set_buffer_uptodate(bh);
946 } else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
948 ocfs2_should_read_blk(inode, page, block_start) &&
949 (block_start < from || block_end > to)) {
950 ll_rw_block(READ, 1, &bh);
954 *p_blkno = *p_blkno + 1;
958 * If we issued read requests - let them complete.
960 while(wait_bh > wait) {
961 wait_on_buffer(*--wait_bh);
962 if (!buffer_uptodate(*wait_bh))
966 if (ret == 0 || !new)
970 * If we get -EIO above, zero out any newly allocated blocks
971 * to avoid exposing stale data.
976 block_end = block_start + bsize;
977 if (block_end <= from)
979 if (block_start >= to)
982 zero_user(page, block_start, bh->b_size);
983 set_buffer_uptodate(bh);
984 mark_buffer_dirty(bh);
987 block_start = block_end;
988 bh = bh->b_this_page;
989 } while (bh != head);
994 #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
995 #define OCFS2_MAX_CTXT_PAGES 1
997 #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
1000 #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
1003 * Describe the state of a single cluster to be written to.
1005 struct ocfs2_write_cluster_desc {
1009 * Give this a unique field because c_phys eventually gets
1013 unsigned c_unwritten;
1014 unsigned c_needs_zero;
1017 struct ocfs2_write_ctxt {
1018 /* Logical cluster position / len of write */
1022 /* First cluster allocated in a nonsparse extend */
1023 u32 w_first_new_cpos;
1025 struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
1028 * This is true if page_size > cluster_size.
1030 * It triggers a set of special cases during write which might
1031 * have to deal with allocating writes to partial pages.
1033 unsigned int w_large_pages;
1036 * Pages involved in this write.
1038 * w_target_page is the page being written to by the user.
1040 * w_pages is an array of pages which always contains
1041 * w_target_page, and in the case of an allocating write with
1042 * page_size < cluster size, it will contain zero'd and mapped
1043 * pages adjacent to w_target_page which need to be written
1044 * out in so that future reads from that region will get
1047 unsigned int w_num_pages;
1048 struct page *w_pages[OCFS2_MAX_CTXT_PAGES];
1049 struct page *w_target_page;
1052 * w_target_locked is used for page_mkwrite path indicating no unlocking
1053 * against w_target_page in ocfs2_write_end_nolock.
1055 unsigned int w_target_locked:1;
1058 * ocfs2_write_end() uses this to know what the real range to
1059 * write in the target should be.
1061 unsigned int w_target_from;
1062 unsigned int w_target_to;
1065 * We could use journal_current_handle() but this is cleaner,
1070 struct buffer_head *w_di_bh;
1072 struct ocfs2_cached_dealloc_ctxt w_dealloc;
1075 void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
1079 for(i = 0; i < num_pages; i++) {
1081 unlock_page(pages[i]);
1082 mark_page_accessed(pages[i]);
1083 page_cache_release(pages[i]);
1088 static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
1093 * w_target_locked is only set to true in the page_mkwrite() case.
1094 * The intent is to allow us to lock the target page from write_begin()
1095 * to write_end(). The caller must hold a ref on w_target_page.
1097 if (wc->w_target_locked) {
1098 BUG_ON(!wc->w_target_page);
1099 for (i = 0; i < wc->w_num_pages; i++) {
1100 if (wc->w_target_page == wc->w_pages[i]) {
1101 wc->w_pages[i] = NULL;
1105 mark_page_accessed(wc->w_target_page);
1106 page_cache_release(wc->w_target_page);
1108 ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
1111 static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
1113 ocfs2_unlock_pages(wc);
1114 brelse(wc->w_di_bh);
1118 static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
1119 struct ocfs2_super *osb, loff_t pos,
1120 unsigned len, struct buffer_head *di_bh)
1123 struct ocfs2_write_ctxt *wc;
1125 wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
1129 wc->w_cpos = pos >> osb->s_clustersize_bits;
1130 wc->w_first_new_cpos = UINT_MAX;
1131 cend = (pos + len - 1) >> osb->s_clustersize_bits;
1132 wc->w_clen = cend - wc->w_cpos + 1;
1134 wc->w_di_bh = di_bh;
1136 if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
1137 wc->w_large_pages = 1;
1139 wc->w_large_pages = 0;
1141 ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
1149 * If a page has any new buffers, zero them out here, and mark them uptodate
1150 * and dirty so they'll be written out (in order to prevent uninitialised
1151 * block data from leaking). And clear the new bit.
1153 static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1155 unsigned int block_start, block_end;
1156 struct buffer_head *head, *bh;
1158 BUG_ON(!PageLocked(page));
1159 if (!page_has_buffers(page))
1162 bh = head = page_buffers(page);
1165 block_end = block_start + bh->b_size;
1167 if (buffer_new(bh)) {
1168 if (block_end > from && block_start < to) {
1169 if (!PageUptodate(page)) {
1170 unsigned start, end;
1172 start = max(from, block_start);
1173 end = min(to, block_end);
1175 zero_user_segment(page, start, end);
1176 set_buffer_uptodate(bh);
1179 clear_buffer_new(bh);
1180 mark_buffer_dirty(bh);
1184 block_start = block_end;
1185 bh = bh->b_this_page;
1186 } while (bh != head);
1190 * Only called when we have a failure during allocating write to write
1191 * zero's to the newly allocated region.
1193 static void ocfs2_write_failure(struct inode *inode,
1194 struct ocfs2_write_ctxt *wc,
1195 loff_t user_pos, unsigned user_len)
1198 unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
1199 to = user_pos + user_len;
1200 struct page *tmppage;
1202 ocfs2_zero_new_buffers(wc->w_target_page, from, to);
1204 for(i = 0; i < wc->w_num_pages; i++) {
1205 tmppage = wc->w_pages[i];
1207 if (page_has_buffers(tmppage)) {
1208 if (ocfs2_should_order_data(inode))
1209 ocfs2_jbd2_file_inode(wc->w_handle, inode);
1211 block_commit_write(tmppage, from, to);
1216 static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
1217 struct ocfs2_write_ctxt *wc,
1218 struct page *page, u32 cpos,
1219 loff_t user_pos, unsigned user_len,
1223 unsigned int map_from = 0, map_to = 0;
1224 unsigned int cluster_start, cluster_end;
1225 unsigned int user_data_from = 0, user_data_to = 0;
1227 ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
1228 &cluster_start, &cluster_end);
1230 /* treat the write as new if the a hole/lseek spanned across
1231 * the page boundary.
1233 new = new | ((i_size_read(inode) <= page_offset(page)) &&
1234 (page_offset(page) <= user_pos));
1236 if (page == wc->w_target_page) {
1237 map_from = user_pos & (PAGE_CACHE_SIZE - 1);
1238 map_to = map_from + user_len;
1241 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1242 cluster_start, cluster_end,
1245 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1246 map_from, map_to, new);
1252 user_data_from = map_from;
1253 user_data_to = map_to;
1255 map_from = cluster_start;
1256 map_to = cluster_end;
1260 * If we haven't allocated the new page yet, we
1261 * shouldn't be writing it out without copying user
1262 * data. This is likely a math error from the caller.
1266 map_from = cluster_start;
1267 map_to = cluster_end;
1269 ret = ocfs2_map_page_blocks(page, p_blkno, inode,
1270 cluster_start, cluster_end, new);
1278 * Parts of newly allocated pages need to be zero'd.
1280 * Above, we have also rewritten 'to' and 'from' - as far as
1281 * the rest of the function is concerned, the entire cluster
1282 * range inside of a page needs to be written.
1284 * We can skip this if the page is up to date - it's already
1285 * been zero'd from being read in as a hole.
1287 if (new && !PageUptodate(page))
1288 ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
1289 cpos, user_data_from, user_data_to);
1291 flush_dcache_page(page);
1298 * This function will only grab one clusters worth of pages.
1300 static int ocfs2_grab_pages_for_write(struct address_space *mapping,
1301 struct ocfs2_write_ctxt *wc,
1302 u32 cpos, loff_t user_pos,
1303 unsigned user_len, int new,
1304 struct page *mmap_page)
1307 unsigned long start, target_index, end_index, index;
1308 struct inode *inode = mapping->host;
1311 target_index = user_pos >> PAGE_CACHE_SHIFT;
1314 * Figure out how many pages we'll be manipulating here. For
1315 * non allocating write, we just change the one
1316 * page. Otherwise, we'll need a whole clusters worth. If we're
1317 * writing past i_size, we only need enough pages to cover the
1318 * last page of the write.
1321 wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
1322 start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
1324 * We need the index *past* the last page we could possibly
1325 * touch. This is the page past the end of the write or
1326 * i_size, whichever is greater.
1328 last_byte = max(user_pos + user_len, i_size_read(inode));
1329 BUG_ON(last_byte < 1);
1330 end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1;
1331 if ((start + wc->w_num_pages) > end_index)
1332 wc->w_num_pages = end_index - start;
1334 wc->w_num_pages = 1;
1335 start = target_index;
1338 for(i = 0; i < wc->w_num_pages; i++) {
1341 if (index == target_index && mmap_page) {
1343 * ocfs2_pagemkwrite() is a little different
1344 * and wants us to directly use the page
1347 lock_page(mmap_page);
1349 /* Exit and let the caller retry */
1350 if (mmap_page->mapping != mapping) {
1351 WARN_ON(mmap_page->mapping);
1352 unlock_page(mmap_page);
1357 page_cache_get(mmap_page);
1358 wc->w_pages[i] = mmap_page;
1359 wc->w_target_locked = true;
1361 wc->w_pages[i] = find_or_create_page(mapping, index,
1363 if (!wc->w_pages[i]) {
1369 wait_for_stable_page(wc->w_pages[i]);
1371 if (index == target_index)
1372 wc->w_target_page = wc->w_pages[i];
1376 wc->w_target_locked = false;
1381 * Prepare a single cluster for write one cluster into the file.
1383 static int ocfs2_write_cluster(struct address_space *mapping,
1384 u32 phys, unsigned int unwritten,
1385 unsigned int should_zero,
1386 struct ocfs2_alloc_context *data_ac,
1387 struct ocfs2_alloc_context *meta_ac,
1388 struct ocfs2_write_ctxt *wc, u32 cpos,
1389 loff_t user_pos, unsigned user_len)
1392 u64 v_blkno, p_blkno;
1393 struct inode *inode = mapping->host;
1394 struct ocfs2_extent_tree et;
1396 new = phys == 0 ? 1 : 0;
1401 * This is safe to call with the page locks - it won't take
1402 * any additional semaphores or cluster locks.
1405 ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
1406 &tmp_pos, 1, 0, wc->w_di_bh,
1407 wc->w_handle, data_ac,
1410 * This shouldn't happen because we must have already
1411 * calculated the correct meta data allocation required. The
1412 * internal tree allocation code should know how to increase
1413 * transaction credits itself.
1415 * If need be, we could handle -EAGAIN for a
1416 * RESTART_TRANS here.
1418 mlog_bug_on_msg(ret == -EAGAIN,
1419 "Inode %llu: EAGAIN return during allocation.\n",
1420 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1425 } else if (unwritten) {
1426 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1428 ret = ocfs2_mark_extent_written(inode, &et,
1429 wc->w_handle, cpos, 1, phys,
1430 meta_ac, &wc->w_dealloc);
1438 v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
1440 v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
1443 * The only reason this should fail is due to an inability to
1444 * find the extent added.
1446 ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
1449 mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
1450 "at logical block %llu",
1451 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1452 (unsigned long long)v_blkno);
1456 BUG_ON(p_blkno == 0);
1458 for(i = 0; i < wc->w_num_pages; i++) {
1461 tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
1462 wc->w_pages[i], cpos,
1473 * We only have cleanup to do in case of allocating write.
1476 ocfs2_write_failure(inode, wc, user_pos, user_len);
1483 static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
1484 struct ocfs2_alloc_context *data_ac,
1485 struct ocfs2_alloc_context *meta_ac,
1486 struct ocfs2_write_ctxt *wc,
1487 loff_t pos, unsigned len)
1491 unsigned int local_len = len;
1492 struct ocfs2_write_cluster_desc *desc;
1493 struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
1495 for (i = 0; i < wc->w_clen; i++) {
1496 desc = &wc->w_desc[i];
1499 * We have to make sure that the total write passed in
1500 * doesn't extend past a single cluster.
1503 cluster_off = pos & (osb->s_clustersize - 1);
1504 if ((cluster_off + local_len) > osb->s_clustersize)
1505 local_len = osb->s_clustersize - cluster_off;
1507 ret = ocfs2_write_cluster(mapping, desc->c_phys,
1511 wc, desc->c_cpos, pos, local_len);
1527 * ocfs2_write_end() wants to know which parts of the target page it
1528 * should complete the write on. It's easiest to compute them ahead of
1529 * time when a more complete view of the write is available.
1531 static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
1532 struct ocfs2_write_ctxt *wc,
1533 loff_t pos, unsigned len, int alloc)
1535 struct ocfs2_write_cluster_desc *desc;
1537 wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
1538 wc->w_target_to = wc->w_target_from + len;
1544 * Allocating write - we may have different boundaries based
1545 * on page size and cluster size.
1547 * NOTE: We can no longer compute one value from the other as
1548 * the actual write length and user provided length may be
1552 if (wc->w_large_pages) {
1554 * We only care about the 1st and last cluster within
1555 * our range and whether they should be zero'd or not. Either
1556 * value may be extended out to the start/end of a
1557 * newly allocated cluster.
1559 desc = &wc->w_desc[0];
1560 if (desc->c_needs_zero)
1561 ocfs2_figure_cluster_boundaries(osb,
1566 desc = &wc->w_desc[wc->w_clen - 1];
1567 if (desc->c_needs_zero)
1568 ocfs2_figure_cluster_boundaries(osb,
1573 wc->w_target_from = 0;
1574 wc->w_target_to = PAGE_CACHE_SIZE;
1579 * Populate each single-cluster write descriptor in the write context
1580 * with information about the i/o to be done.
1582 * Returns the number of clusters that will have to be allocated, as
1583 * well as a worst case estimate of the number of extent records that
1584 * would have to be created during a write to an unwritten region.
1586 static int ocfs2_populate_write_desc(struct inode *inode,
1587 struct ocfs2_write_ctxt *wc,
1588 unsigned int *clusters_to_alloc,
1589 unsigned int *extents_to_split)
1592 struct ocfs2_write_cluster_desc *desc;
1593 unsigned int num_clusters = 0;
1594 unsigned int ext_flags = 0;
1598 *clusters_to_alloc = 0;
1599 *extents_to_split = 0;
1601 for (i = 0; i < wc->w_clen; i++) {
1602 desc = &wc->w_desc[i];
1603 desc->c_cpos = wc->w_cpos + i;
1605 if (num_clusters == 0) {
1607 * Need to look up the next extent record.
1609 ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
1610 &num_clusters, &ext_flags);
1616 /* We should already CoW the refcountd extent. */
1617 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1620 * Assume worst case - that we're writing in
1621 * the middle of the extent.
1623 * We can assume that the write proceeds from
1624 * left to right, in which case the extent
1625 * insert code is smart enough to coalesce the
1626 * next splits into the previous records created.
1628 if (ext_flags & OCFS2_EXT_UNWRITTEN)
1629 *extents_to_split = *extents_to_split + 2;
1632 * Only increment phys if it doesn't describe
1639 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
1640 * file that got extended. w_first_new_cpos tells us
1641 * where the newly allocated clusters are so we can
1644 if (desc->c_cpos >= wc->w_first_new_cpos) {
1646 desc->c_needs_zero = 1;
1649 desc->c_phys = phys;
1652 desc->c_needs_zero = 1;
1653 *clusters_to_alloc = *clusters_to_alloc + 1;
1656 if (ext_flags & OCFS2_EXT_UNWRITTEN) {
1657 desc->c_unwritten = 1;
1658 desc->c_needs_zero = 1;
1669 static int ocfs2_write_begin_inline(struct address_space *mapping,
1670 struct inode *inode,
1671 struct ocfs2_write_ctxt *wc)
1674 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1677 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1679 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1680 if (IS_ERR(handle)) {
1681 ret = PTR_ERR(handle);
1686 page = find_or_create_page(mapping, 0, GFP_NOFS);
1688 ocfs2_commit_trans(osb, handle);
1694 * If we don't set w_num_pages then this page won't get unlocked
1695 * and freed on cleanup of the write context.
1697 wc->w_pages[0] = wc->w_target_page = page;
1698 wc->w_num_pages = 1;
1700 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
1701 OCFS2_JOURNAL_ACCESS_WRITE);
1703 ocfs2_commit_trans(osb, handle);
1709 if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1710 ocfs2_set_inode_data_inline(inode, di);
1712 if (!PageUptodate(page)) {
1713 ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
1715 ocfs2_commit_trans(osb, handle);
1721 wc->w_handle = handle;
1726 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
1728 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1730 if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
1735 static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
1736 struct inode *inode, loff_t pos,
1737 unsigned len, struct page *mmap_page,
1738 struct ocfs2_write_ctxt *wc)
1740 int ret, written = 0;
1741 loff_t end = pos + len;
1742 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1743 struct ocfs2_dinode *di = NULL;
1745 trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
1746 len, (unsigned long long)pos,
1747 oi->ip_dyn_features);
1750 * Handle inodes which already have inline data 1st.
1752 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1753 if (mmap_page == NULL &&
1754 ocfs2_size_fits_inline_data(wc->w_di_bh, end))
1755 goto do_inline_write;
1758 * The write won't fit - we have to give this inode an
1759 * inline extent list now.
1761 ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
1768 * Check whether the inode can accept inline data.
1770 if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
1774 * Check whether the write can fit.
1776 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1778 end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
1782 ret = ocfs2_write_begin_inline(mapping, inode, wc);
1789 * This signals to the caller that the data can be written
1794 return written ? written : ret;
1798 * This function only does anything for file systems which can't
1799 * handle sparse files.
1801 * What we want to do here is fill in any hole between the current end
1802 * of allocation and the end of our write. That way the rest of the
1803 * write path can treat it as an non-allocating write, which has no
1804 * special case code for sparse/nonsparse files.
1806 static int ocfs2_expand_nonsparse_inode(struct inode *inode,
1807 struct buffer_head *di_bh,
1808 loff_t pos, unsigned len,
1809 struct ocfs2_write_ctxt *wc)
1812 loff_t newsize = pos + len;
1814 BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1816 if (newsize <= i_size_read(inode))
1819 ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
1823 wc->w_first_new_cpos =
1824 ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
1829 static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
1834 BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
1835 if (pos > i_size_read(inode))
1836 ret = ocfs2_zero_extend(inode, di_bh, pos);
1842 * Try to flush truncate logs if we can free enough clusters from it.
1843 * As for return value, "< 0" means error, "0" no space and "1" means
1844 * we have freed enough spaces and let the caller try to allocate again.
1846 static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
1847 unsigned int needed)
1851 unsigned int truncated_clusters;
1853 mutex_lock(&osb->osb_tl_inode->i_mutex);
1854 truncated_clusters = osb->truncated_clusters;
1855 mutex_unlock(&osb->osb_tl_inode->i_mutex);
1858 * Check whether we can succeed in allocating if we free
1861 if (truncated_clusters < needed)
1864 ret = ocfs2_flush_truncate_log(osb);
1870 if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
1871 jbd2_log_wait_commit(osb->journal->j_journal, target);
1878 int ocfs2_write_begin_nolock(struct file *filp,
1879 struct address_space *mapping,
1880 loff_t pos, unsigned len, unsigned flags,
1881 struct page **pagep, void **fsdata,
1882 struct buffer_head *di_bh, struct page *mmap_page)
1884 int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
1885 unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
1886 struct ocfs2_write_ctxt *wc;
1887 struct inode *inode = mapping->host;
1888 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1889 struct ocfs2_dinode *di;
1890 struct ocfs2_alloc_context *data_ac = NULL;
1891 struct ocfs2_alloc_context *meta_ac = NULL;
1893 struct ocfs2_extent_tree et;
1894 int try_free = 1, ret1;
1897 ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
1903 if (ocfs2_supports_inline_data(osb)) {
1904 ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
1916 if (ocfs2_sparse_alloc(osb))
1917 ret = ocfs2_zero_tail(inode, di_bh, pos);
1919 ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len,
1926 ret = ocfs2_check_range_for_refcount(inode, pos, len);
1930 } else if (ret == 1) {
1931 clusters_need = wc->w_clen;
1932 ret = ocfs2_refcount_cow(inode, di_bh,
1933 wc->w_cpos, wc->w_clen, UINT_MAX);
1940 ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
1946 clusters_need += clusters_to_alloc;
1948 di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
1950 trace_ocfs2_write_begin_nolock(
1951 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1952 (long long)i_size_read(inode),
1953 le32_to_cpu(di->i_clusters),
1954 pos, len, flags, mmap_page,
1955 clusters_to_alloc, extents_to_split);
1958 * We set w_target_from, w_target_to here so that
1959 * ocfs2_write_end() knows which range in the target page to
1960 * write out. An allocation requires that we write the entire
1963 if (clusters_to_alloc || extents_to_split) {
1965 * XXX: We are stretching the limits of
1966 * ocfs2_lock_allocators(). It greatly over-estimates
1967 * the work to be done.
1969 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
1971 ret = ocfs2_lock_allocators(inode, &et,
1972 clusters_to_alloc, extents_to_split,
1973 &data_ac, &meta_ac);
1980 data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
1982 credits = ocfs2_calc_extend_credits(inode->i_sb,
1988 * We have to zero sparse allocated clusters, unwritten extent clusters,
1989 * and non-sparse clusters we just extended. For non-sparse writes,
1990 * we know zeros will only be needed in the first and/or last cluster.
1992 if (clusters_to_alloc || extents_to_split ||
1993 (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
1994 wc->w_desc[wc->w_clen - 1].c_needs_zero)))
1995 cluster_of_pages = 1;
1997 cluster_of_pages = 0;
1999 ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
2001 handle = ocfs2_start_trans(osb, credits);
2002 if (IS_ERR(handle)) {
2003 ret = PTR_ERR(handle);
2008 wc->w_handle = handle;
2010 if (clusters_to_alloc) {
2011 ret = dquot_alloc_space_nodirty(inode,
2012 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
2017 * We don't want this to fail in ocfs2_write_end(), so do it
2020 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
2021 OCFS2_JOURNAL_ACCESS_WRITE);
2028 * Fill our page array first. That way we've grabbed enough so
2029 * that we can zero and flush if we error after adding the
2032 ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
2033 cluster_of_pages, mmap_page);
2034 if (ret && ret != -EAGAIN) {
2040 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
2041 * the target page. In this case, we exit with no error and no target
2042 * page. This will trigger the caller, page_mkwrite(), to re-try
2045 if (ret == -EAGAIN) {
2046 BUG_ON(wc->w_target_page);
2051 ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
2059 ocfs2_free_alloc_context(data_ac);
2061 ocfs2_free_alloc_context(meta_ac);
2064 *pagep = wc->w_target_page;
2068 if (clusters_to_alloc)
2069 dquot_free_space(inode,
2070 ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
2072 ocfs2_commit_trans(osb, handle);
2075 ocfs2_free_write_ctxt(wc);
2078 ocfs2_free_alloc_context(data_ac);
2082 ocfs2_free_alloc_context(meta_ac);
2086 if (ret == -ENOSPC && try_free) {
2088 * Try to free some truncate log so that we can have enough
2089 * clusters to allocate.
2093 ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
2104 static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
2105 loff_t pos, unsigned len, unsigned flags,
2106 struct page **pagep, void **fsdata)
2109 struct buffer_head *di_bh = NULL;
2110 struct inode *inode = mapping->host;
2112 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2119 * Take alloc sem here to prevent concurrent lookups. That way
2120 * the mapping, zeroing and tree manipulation within
2121 * ocfs2_write() will be safe against ->readpage(). This
2122 * should also serve to lock out allocation from a shared
2125 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2127 ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep,
2128 fsdata, di_bh, NULL);
2139 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2142 ocfs2_inode_unlock(inode, 1);
2147 static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
2148 unsigned len, unsigned *copied,
2149 struct ocfs2_dinode *di,
2150 struct ocfs2_write_ctxt *wc)
2154 if (unlikely(*copied < len)) {
2155 if (!PageUptodate(wc->w_target_page)) {
2161 kaddr = kmap_atomic(wc->w_target_page);
2162 memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
2163 kunmap_atomic(kaddr);
2165 trace_ocfs2_write_end_inline(
2166 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2167 (unsigned long long)pos, *copied,
2168 le16_to_cpu(di->id2.i_data.id_count),
2169 le16_to_cpu(di->i_dyn_features));
2172 int ocfs2_write_end_nolock(struct address_space *mapping,
2173 loff_t pos, unsigned len, unsigned copied,
2174 struct page *page, void *fsdata)
2177 unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
2178 struct inode *inode = mapping->host;
2179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2180 struct ocfs2_write_ctxt *wc = fsdata;
2181 struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
2182 handle_t *handle = wc->w_handle;
2183 struct page *tmppage;
2185 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2186 ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
2187 goto out_write_size;
2190 if (unlikely(copied < len)) {
2191 if (!PageUptodate(wc->w_target_page))
2194 ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
2197 flush_dcache_page(wc->w_target_page);
2199 for(i = 0; i < wc->w_num_pages; i++) {
2200 tmppage = wc->w_pages[i];
2202 if (tmppage == wc->w_target_page) {
2203 from = wc->w_target_from;
2204 to = wc->w_target_to;
2206 BUG_ON(from > PAGE_CACHE_SIZE ||
2207 to > PAGE_CACHE_SIZE ||
2211 * Pages adjacent to the target (if any) imply
2212 * a hole-filling write in which case we want
2213 * to flush their entire range.
2216 to = PAGE_CACHE_SIZE;
2219 if (page_has_buffers(tmppage)) {
2220 if (ocfs2_should_order_data(inode))
2221 ocfs2_jbd2_file_inode(wc->w_handle, inode);
2222 block_commit_write(tmppage, from, to);
2228 if (pos > i_size_read(inode)) {
2229 i_size_write(inode, pos);
2230 mark_inode_dirty(inode);
2232 inode->i_blocks = ocfs2_inode_sector_count(inode);
2233 di->i_size = cpu_to_le64((u64)i_size_read(inode));
2234 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2235 di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
2236 di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
2237 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2238 ocfs2_journal_dirty(handle, wc->w_di_bh);
2240 /* unlock pages before dealloc since it needs acquiring j_trans_barrier
2241 * lock, or it will cause a deadlock since journal commit threads holds
2242 * this lock and will ask for the page lock when flushing the data.
2243 * put it here to preserve the unlock order.
2245 ocfs2_unlock_pages(wc);
2247 ocfs2_commit_trans(osb, handle);
2249 ocfs2_run_deallocs(osb, &wc->w_dealloc);
2251 brelse(wc->w_di_bh);
2257 static int ocfs2_write_end(struct file *file, struct address_space *mapping,
2258 loff_t pos, unsigned len, unsigned copied,
2259 struct page *page, void *fsdata)
2262 struct inode *inode = mapping->host;
2264 ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
2266 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2267 ocfs2_inode_unlock(inode, 1);
2272 const struct address_space_operations ocfs2_aops = {
2273 .readpage = ocfs2_readpage,
2274 .readpages = ocfs2_readpages,
2275 .writepage = ocfs2_writepage,
2276 .write_begin = ocfs2_write_begin,
2277 .write_end = ocfs2_write_end,
2279 .direct_IO = ocfs2_direct_IO,
2280 .invalidatepage = block_invalidatepage,
2281 .releasepage = ocfs2_releasepage,
2282 .migratepage = buffer_migrate_page,
2283 .is_partially_uptodate = block_is_partially_uptodate,
2284 .error_remove_page = generic_error_remove_page,