2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/pagemap.h>
16 #include <linux/pagevec.h>
17 #include <linux/mpage.h>
19 #include <linux/writeback.h>
20 #include <linux/swap.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/lm_interface.h>
23 #include <linux/backing-dev.h>
24 #include <linux/pagevec.h>
33 #include "ops_address.h"
42 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
43 unsigned int from, unsigned int to)
45 struct buffer_head *head = page_buffers(page);
46 unsigned int bsize = head->b_size;
47 struct buffer_head *bh;
48 unsigned int start, end;
50 for (bh = head, start = 0; bh != head || !start;
51 bh = bh->b_this_page, start = end) {
53 if (end <= from || start >= to)
55 if (gfs2_is_jdata(ip))
56 set_buffer_uptodate(bh);
57 gfs2_trans_add_bh(ip->i_gl, bh, 0);
62 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
64 * @lblock: The block number to look up
65 * @bh_result: The buffer head to return the result in
66 * @create: Non-zero if we may add block to the file
71 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
72 struct buffer_head *bh_result, int create)
76 error = gfs2_block_map(inode, lblock, bh_result, 0);
79 if (!buffer_mapped(bh_result))
84 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
85 struct buffer_head *bh_result, int create)
87 return gfs2_block_map(inode, lblock, bh_result, 0);
91 * gfs2_writepage_common - Common bits of writepage
92 * @page: The page to be written
93 * @wbc: The writeback control
95 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
98 static int gfs2_writepage_common(struct page *page,
99 struct writeback_control *wbc)
101 struct inode *inode = page->mapping->host;
102 struct gfs2_inode *ip = GFS2_I(inode);
103 struct gfs2_sbd *sdp = GFS2_SB(inode);
104 loff_t i_size = i_size_read(inode);
105 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
108 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
110 if (current->journal_info)
112 /* Is the page fully outside i_size? (truncate in progress) */
113 offset = i_size & (PAGE_CACHE_SIZE-1);
114 if (page->index > end_index || (page->index == end_index && !offset)) {
115 page->mapping->a_ops->invalidatepage(page, 0);
120 redirty_page_for_writepage(wbc, page);
127 * gfs2_writeback_writepage - Write page for writeback mappings
129 * @wbc: The writeback control
133 static int gfs2_writeback_writepage(struct page *page,
134 struct writeback_control *wbc)
138 ret = gfs2_writepage_common(page, wbc);
142 ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
144 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
149 * gfs2_ordered_writepage - Write page for ordered data files
150 * @page: The page to write
151 * @wbc: The writeback control
155 static int gfs2_ordered_writepage(struct page *page,
156 struct writeback_control *wbc)
158 struct inode *inode = page->mapping->host;
159 struct gfs2_inode *ip = GFS2_I(inode);
162 ret = gfs2_writepage_common(page, wbc);
166 if (!page_has_buffers(page)) {
167 create_empty_buffers(page, inode->i_sb->s_blocksize,
168 (1 << BH_Dirty)|(1 << BH_Uptodate));
170 gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
171 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
175 * __gfs2_jdata_writepage - The core of jdata writepage
176 * @page: The page to write
177 * @wbc: The writeback control
179 * This is shared between writepage and writepages and implements the
180 * core of the writepage operation. If a transaction is required then
181 * PageChecked will have been set and the transaction will have
182 * already been started before this is called.
185 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
187 struct inode *inode = page->mapping->host;
188 struct gfs2_inode *ip = GFS2_I(inode);
189 struct gfs2_sbd *sdp = GFS2_SB(inode);
191 if (PageChecked(page)) {
192 ClearPageChecked(page);
193 if (!page_has_buffers(page)) {
194 create_empty_buffers(page, inode->i_sb->s_blocksize,
195 (1 << BH_Dirty)|(1 << BH_Uptodate));
197 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
199 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
203 * gfs2_jdata_writepage - Write complete page
204 * @page: Page to write
210 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
212 struct inode *inode = page->mapping->host;
213 struct gfs2_sbd *sdp = GFS2_SB(inode);
217 error = gfs2_writepage_common(page, wbc);
221 if (PageChecked(page)) {
222 if (wbc->sync_mode != WB_SYNC_ALL)
224 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
229 error = __gfs2_jdata_writepage(page, wbc);
235 redirty_page_for_writepage(wbc, page);
241 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
242 * @mapping: The mapping to write
243 * @wbc: Write-back control
245 * For the data=writeback case we can already ignore buffer heads
246 * and write whole extents at once. This is a big reduction in the
247 * number of I/O requests we send and the bmap calls we make in this case.
249 static int gfs2_writeback_writepages(struct address_space *mapping,
250 struct writeback_control *wbc)
252 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
256 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
257 * @mapping: The mapping
258 * @wbc: The writeback control
259 * @writepage: The writepage function to call for each page
260 * @pvec: The vector of pages
261 * @nr_pages: The number of pages to write
263 * Returns: non-zero if loop should terminate, zero otherwise
266 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
267 struct writeback_control *wbc,
268 struct pagevec *pvec,
269 int nr_pages, pgoff_t end)
271 struct inode *inode = mapping->host;
272 struct gfs2_sbd *sdp = GFS2_SB(inode);
273 loff_t i_size = i_size_read(inode);
274 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
275 unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
276 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
277 struct backing_dev_info *bdi = mapping->backing_dev_info;
281 ret = gfs2_trans_begin(sdp, nrblocks, 0);
285 for(i = 0; i < nr_pages; i++) {
286 struct page *page = pvec->pages[i];
290 if (unlikely(page->mapping != mapping)) {
295 if (!wbc->range_cyclic && page->index > end) {
301 if (wbc->sync_mode != WB_SYNC_NONE)
302 wait_on_page_writeback(page);
304 if (PageWriteback(page) ||
305 !clear_page_dirty_for_io(page)) {
310 /* Is the page fully outside i_size? (truncate in progress) */
311 if (page->index > end_index || (page->index == end_index && !offset)) {
312 page->mapping->a_ops->invalidatepage(page, 0);
317 ret = __gfs2_jdata_writepage(page, wbc);
319 if (ret || (--(wbc->nr_to_write) <= 0))
321 if (wbc->nonblocking && bdi_write_congested(bdi)) {
322 wbc->encountered_congestion = 1;
332 * gfs2_write_cache_jdata - Like write_cache_pages but different
333 * @mapping: The mapping to write
334 * @wbc: The writeback control
335 * @writepage: The writepage function to call
336 * @data: The data to pass to writepage
338 * The reason that we use our own function here is that we need to
339 * start transactions before we grab page locks. This allows us
340 * to get the ordering right.
343 static int gfs2_write_cache_jdata(struct address_space *mapping,
344 struct writeback_control *wbc)
346 struct backing_dev_info *bdi = mapping->backing_dev_info;
356 if (wbc->nonblocking && bdi_write_congested(bdi)) {
357 wbc->encountered_congestion = 1;
361 pagevec_init(&pvec, 0);
362 if (wbc->range_cyclic) {
363 index = mapping->writeback_index; /* Start from prev offset */
366 index = wbc->range_start >> PAGE_CACHE_SHIFT;
367 end = wbc->range_end >> PAGE_CACHE_SHIFT;
368 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
374 while (!done && (index <= end) &&
375 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
377 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
379 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
385 pagevec_release(&pvec);
389 if (!scanned && !done) {
391 * We hit the last page and there is more work to be done: wrap
392 * back to the start of the file
399 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
400 mapping->writeback_index = index;
406 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
407 * @mapping: The mapping to write
408 * @wbc: The writeback control
412 static int gfs2_jdata_writepages(struct address_space *mapping,
413 struct writeback_control *wbc)
415 struct gfs2_inode *ip = GFS2_I(mapping->host);
416 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
419 ret = gfs2_write_cache_jdata(mapping, wbc);
420 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
421 gfs2_log_flush(sdp, ip->i_gl);
422 ret = gfs2_write_cache_jdata(mapping, wbc);
428 * stuffed_readpage - Fill in a Linux page with stuffed file data
435 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
437 struct buffer_head *dibh;
442 * Due to the order of unstuffing files and ->nopage(), we can be
443 * asked for a zero page in the case of a stuffed file being extended,
444 * so we need to supply one here. It doesn't happen often.
446 if (unlikely(page->index)) {
447 zero_user(page, 0, PAGE_CACHE_SIZE);
451 error = gfs2_meta_inode_buffer(ip, &dibh);
455 kaddr = kmap_atomic(page, KM_USER0);
456 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
458 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
459 kunmap_atomic(kaddr, KM_USER0);
460 flush_dcache_page(page);
462 SetPageUptodate(page);
469 * __gfs2_readpage - readpage
470 * @file: The file to read a page for
471 * @page: The page to read
473 * This is the core of gfs2's readpage. Its used by the internal file
474 * reading code as in that case we already hold the glock. Also its
475 * called by gfs2_readpage() once the required lock has been granted.
479 static int __gfs2_readpage(void *file, struct page *page)
481 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
482 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
485 if (gfs2_is_stuffed(ip)) {
486 error = stuffed_readpage(ip, page);
489 error = mpage_readpage(page, gfs2_block_map);
492 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
499 * gfs2_readpage - read a page of a file
500 * @file: The file to read
501 * @page: The page of the file
503 * This deals with the locking required. We use a trylock in order to
504 * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
505 * in the event that we are unable to get the lock.
508 static int gfs2_readpage(struct file *file, struct page *page)
510 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
511 struct gfs2_holder *gh;
514 gh = gfs2_glock_is_locked_by_me(ip->i_gl);
516 gh = kmalloc(sizeof(struct gfs2_holder), GFP_NOFS);
519 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, gh);
521 error = gfs2_glock_nq_atime(gh);
522 if (likely(error != 0))
524 return AOP_TRUNCATED_PAGE;
526 error = __gfs2_readpage(file, page);
529 gfs2_holder_uninit(gh);
535 * gfs2_internal_read - read an internal file
536 * @ip: The gfs2 inode
537 * @ra_state: The readahead state (or NULL for no readahead)
538 * @buf: The buffer to fill
539 * @pos: The file position
540 * @size: The amount to read
544 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
545 char *buf, loff_t *pos, unsigned size)
547 struct address_space *mapping = ip->i_inode.i_mapping;
548 unsigned long index = *pos / PAGE_CACHE_SIZE;
549 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
557 if (offset + size > PAGE_CACHE_SIZE)
558 amt = PAGE_CACHE_SIZE - offset;
559 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
561 return PTR_ERR(page);
562 p = kmap_atomic(page, KM_USER0);
563 memcpy(buf + copied, p + offset, amt);
564 kunmap_atomic(p, KM_USER0);
565 mark_page_accessed(page);
566 page_cache_release(page);
570 } while(copied < size);
576 * gfs2_readpages - Read a bunch of pages at once
579 * 1. This is only for readahead, so we can simply ignore any things
580 * which are slightly inconvenient (such as locking conflicts between
581 * the page lock and the glock) and return having done no I/O. Its
582 * obviously not something we'd want to do on too regular a basis.
583 * Any I/O we ignore at this time will be done via readpage later.
584 * 2. We don't handle stuffed files here we let readpage do the honours.
585 * 3. mpage_readpages() does most of the heavy lifting in the common case.
586 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
589 static int gfs2_readpages(struct file *file, struct address_space *mapping,
590 struct list_head *pages, unsigned nr_pages)
592 struct inode *inode = mapping->host;
593 struct gfs2_inode *ip = GFS2_I(inode);
594 struct gfs2_sbd *sdp = GFS2_SB(inode);
595 struct gfs2_holder gh;
598 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
599 ret = gfs2_glock_nq_atime(&gh);
602 if (!gfs2_is_stuffed(ip))
603 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
606 gfs2_holder_uninit(&gh);
607 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
613 * gfs2_write_begin - Begin to write to a file
614 * @file: The file to write to
615 * @mapping: The mapping in which to write
616 * @pos: The file offset at which to start writing
617 * @len: Length of the write
618 * @flags: Various flags
619 * @pagep: Pointer to return the page
620 * @fsdata: Pointer to return fs data (unused by GFS2)
625 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
626 loff_t pos, unsigned len, unsigned flags,
627 struct page **pagep, void **fsdata)
629 struct gfs2_inode *ip = GFS2_I(mapping->host);
630 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
631 unsigned int data_blocks, ind_blocks, rblocks;
634 struct gfs2_alloc *al;
635 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
636 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
637 unsigned to = from + len;
640 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
641 error = gfs2_glock_nq_atime(&ip->i_gh);
645 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
646 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
650 if (alloc_required) {
651 al = gfs2_alloc_get(ip);
653 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
657 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
661 al->al_requested = data_blocks + ind_blocks;
662 error = gfs2_inplace_reserve(ip);
667 rblocks = RES_DINODE + ind_blocks;
668 if (gfs2_is_jdata(ip))
669 rblocks += data_blocks ? data_blocks : 1;
670 if (ind_blocks || data_blocks)
671 rblocks += RES_STATFS + RES_QUOTA;
673 error = gfs2_trans_begin(sdp, rblocks,
674 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
679 page = __grab_cache_page(mapping, index);
684 if (gfs2_is_stuffed(ip)) {
686 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
687 error = gfs2_unstuff_dinode(ip, page);
690 } else if (!PageUptodate(page)) {
691 error = stuffed_readpage(ip, page);
697 error = block_prepare_write(page, from, to, gfs2_block_map);
702 page_cache_release(page);
703 if (pos + len > ip->i_inode.i_size)
704 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
708 if (alloc_required) {
709 gfs2_inplace_release(ip);
711 gfs2_quota_unlock(ip);
716 gfs2_glock_dq(&ip->i_gh);
718 gfs2_holder_uninit(&ip->i_gh);
723 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
724 * @inode: the rindex inode
726 static void adjust_fs_space(struct inode *inode)
728 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
729 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
730 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
731 u64 fs_total, new_free;
733 /* Total up the file system space, according to the latest rindex. */
734 fs_total = gfs2_ri_total(sdp);
736 spin_lock(&sdp->sd_statfs_spin);
737 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
738 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
741 spin_unlock(&sdp->sd_statfs_spin);
742 fs_warn(sdp, "File system extended by %llu blocks.\n",
743 (unsigned long long)new_free);
744 gfs2_statfs_change(sdp, new_free, new_free, 0);
748 * gfs2_stuffed_write_end - Write end for stuffed files
750 * @dibh: The buffer_head containing the on-disk inode
751 * @pos: The file position
752 * @len: The length of the write
753 * @copied: How much was actually copied by the VFS
756 * This copies the data from the page into the inode block after
757 * the inode data structure itself.
761 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
762 loff_t pos, unsigned len, unsigned copied,
765 struct gfs2_inode *ip = GFS2_I(inode);
766 struct gfs2_sbd *sdp = GFS2_SB(inode);
767 u64 to = pos + copied;
769 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
770 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
772 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
773 kaddr = kmap_atomic(page, KM_USER0);
774 memcpy(buf + pos, kaddr + pos, copied);
775 memset(kaddr + pos + copied, 0, len - copied);
776 flush_dcache_page(page);
777 kunmap_atomic(kaddr, KM_USER0);
779 if (!PageUptodate(page))
780 SetPageUptodate(page);
782 page_cache_release(page);
784 if (inode->i_size < to) {
785 i_size_write(inode, to);
786 ip->i_di.di_size = inode->i_size;
787 di->di_size = cpu_to_be64(inode->i_size);
788 mark_inode_dirty(inode);
791 if (inode == sdp->sd_rindex)
792 adjust_fs_space(inode);
796 gfs2_glock_dq(&ip->i_gh);
797 gfs2_holder_uninit(&ip->i_gh);
803 * @file: The file to write to
804 * @mapping: The address space to write to
805 * @pos: The file position
806 * @len: The length of the data
808 * @page: The page that has been written
809 * @fsdata: The fsdata (unused in GFS2)
811 * The main write_end function for GFS2. We have a separate one for
812 * stuffed files as they are slightly different, otherwise we just
813 * put our locking around the VFS provided functions.
818 static int gfs2_write_end(struct file *file, struct address_space *mapping,
819 loff_t pos, unsigned len, unsigned copied,
820 struct page *page, void *fsdata)
822 struct inode *inode = page->mapping->host;
823 struct gfs2_inode *ip = GFS2_I(inode);
824 struct gfs2_sbd *sdp = GFS2_SB(inode);
825 struct buffer_head *dibh;
826 struct gfs2_alloc *al = ip->i_alloc;
827 struct gfs2_dinode *di;
828 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
829 unsigned int to = from + len;
832 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
834 ret = gfs2_meta_inode_buffer(ip, &dibh);
837 page_cache_release(page);
841 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
843 if (gfs2_is_stuffed(ip))
844 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
846 if (!gfs2_is_writeback(ip))
847 gfs2_page_add_databufs(ip, page, from, to);
849 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
851 if (likely(ret >= 0) && (inode->i_size > ip->i_di.di_size)) {
852 di = (struct gfs2_dinode *)dibh->b_data;
853 ip->i_di.di_size = inode->i_size;
854 di->di_size = cpu_to_be64(inode->i_size);
855 mark_inode_dirty(inode);
858 if (inode == sdp->sd_rindex)
859 adjust_fs_space(inode);
865 gfs2_inplace_release(ip);
866 gfs2_quota_unlock(ip);
869 gfs2_glock_dq(&ip->i_gh);
870 gfs2_holder_uninit(&ip->i_gh);
875 * gfs2_set_page_dirty - Page dirtying function
876 * @page: The page to dirty
878 * Returns: 1 if it dirtyed the page, or 0 otherwise
881 static int gfs2_set_page_dirty(struct page *page)
883 SetPageChecked(page);
884 return __set_page_dirty_buffers(page);
888 * gfs2_bmap - Block map function
889 * @mapping: Address space info
890 * @lblock: The block to map
892 * Returns: The disk address for the block or 0 on hole or error
895 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
897 struct gfs2_inode *ip = GFS2_I(mapping->host);
898 struct gfs2_holder i_gh;
902 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
906 if (!gfs2_is_stuffed(ip))
907 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
909 gfs2_glock_dq_uninit(&i_gh);
914 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
916 struct gfs2_bufdata *bd;
920 clear_buffer_dirty(bh);
923 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
924 list_del_init(&bd->bd_le.le_list);
926 gfs2_remove_from_journal(bh, current->journal_info, 0);
929 clear_buffer_mapped(bh);
930 clear_buffer_req(bh);
931 clear_buffer_new(bh);
932 gfs2_log_unlock(sdp);
936 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
938 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
939 struct buffer_head *bh, *head;
940 unsigned long pos = 0;
942 BUG_ON(!PageLocked(page));
944 ClearPageChecked(page);
945 if (!page_has_buffers(page))
948 bh = head = page_buffers(page);
951 gfs2_discard(sdp, bh);
953 bh = bh->b_this_page;
954 } while (bh != head);
957 try_to_release_page(page, 0);
961 * gfs2_ok_for_dio - check that dio is valid on this file
964 * @offset: The offset at which we are reading or writing
966 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
967 * 1 (to accept the i/o request)
969 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
972 * Should we return an error here? I can't see that O_DIRECT for
973 * a stuffed file makes any sense. For now we'll silently fall
974 * back to buffered I/O
976 if (gfs2_is_stuffed(ip))
979 if (offset > i_size_read(&ip->i_inode))
986 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
987 const struct iovec *iov, loff_t offset,
988 unsigned long nr_segs)
990 struct file *file = iocb->ki_filp;
991 struct inode *inode = file->f_mapping->host;
992 struct gfs2_inode *ip = GFS2_I(inode);
993 struct gfs2_holder gh;
997 * Deferred lock, even if its a write, since we do no allocation
998 * on this path. All we need change is atime, and this lock mode
999 * ensures that other nodes have flushed their buffered read caches
1000 * (i.e. their page cache entries for this inode). We do not,
1001 * unfortunately have the option of only flushing a range like
1004 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
1005 rv = gfs2_glock_nq_atime(&gh);
1008 rv = gfs2_ok_for_dio(ip, rw, offset);
1010 goto out; /* dio not valid, fall back to buffered i/o */
1012 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1013 iov, offset, nr_segs,
1014 gfs2_get_block_direct, NULL);
1016 gfs2_glock_dq_m(1, &gh);
1017 gfs2_holder_uninit(&gh);
1022 * gfs2_releasepage - free the metadata associated with a page
1023 * @page: the page that's being released
1024 * @gfp_mask: passed from Linux VFS, ignored by us
1026 * Call try_to_free_buffers() if the buffers in this page can be
1032 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1034 struct inode *aspace = page->mapping->host;
1035 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1036 struct buffer_head *bh, *head;
1037 struct gfs2_bufdata *bd;
1039 if (!page_has_buffers(page))
1043 head = bh = page_buffers(page);
1045 if (atomic_read(&bh->b_count))
1046 goto cannot_release;
1048 if (bd && bd->bd_ail)
1049 goto cannot_release;
1050 gfs2_assert_warn(sdp, !buffer_pinned(bh));
1051 gfs2_assert_warn(sdp, !buffer_dirty(bh));
1052 bh = bh->b_this_page;
1053 } while(bh != head);
1054 gfs2_log_unlock(sdp);
1056 head = bh = page_buffers(page);
1061 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1062 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
1063 if (!list_empty(&bd->bd_le.le_list)) {
1064 if (!buffer_pinned(bh))
1065 list_del_init(&bd->bd_le.le_list);
1071 bh->b_private = NULL;
1073 gfs2_log_unlock(sdp);
1075 kmem_cache_free(gfs2_bufdata_cachep, bd);
1077 bh = bh->b_this_page;
1078 } while (bh != head);
1080 return try_to_free_buffers(page);
1082 gfs2_log_unlock(sdp);
1086 static const struct address_space_operations gfs2_writeback_aops = {
1087 .writepage = gfs2_writeback_writepage,
1088 .writepages = gfs2_writeback_writepages,
1089 .readpage = gfs2_readpage,
1090 .readpages = gfs2_readpages,
1091 .sync_page = block_sync_page,
1092 .write_begin = gfs2_write_begin,
1093 .write_end = gfs2_write_end,
1095 .invalidatepage = gfs2_invalidatepage,
1096 .releasepage = gfs2_releasepage,
1097 .direct_IO = gfs2_direct_IO,
1098 .migratepage = buffer_migrate_page,
1101 static const struct address_space_operations gfs2_ordered_aops = {
1102 .writepage = gfs2_ordered_writepage,
1103 .readpage = gfs2_readpage,
1104 .readpages = gfs2_readpages,
1105 .sync_page = block_sync_page,
1106 .write_begin = gfs2_write_begin,
1107 .write_end = gfs2_write_end,
1108 .set_page_dirty = gfs2_set_page_dirty,
1110 .invalidatepage = gfs2_invalidatepage,
1111 .releasepage = gfs2_releasepage,
1112 .direct_IO = gfs2_direct_IO,
1113 .migratepage = buffer_migrate_page,
1116 static const struct address_space_operations gfs2_jdata_aops = {
1117 .writepage = gfs2_jdata_writepage,
1118 .writepages = gfs2_jdata_writepages,
1119 .readpage = gfs2_readpage,
1120 .readpages = gfs2_readpages,
1121 .sync_page = block_sync_page,
1122 .write_begin = gfs2_write_begin,
1123 .write_end = gfs2_write_end,
1124 .set_page_dirty = gfs2_set_page_dirty,
1126 .invalidatepage = gfs2_invalidatepage,
1127 .releasepage = gfs2_releasepage,
1130 void gfs2_set_aops(struct inode *inode)
1132 struct gfs2_inode *ip = GFS2_I(inode);
1134 if (gfs2_is_writeback(ip))
1135 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1136 else if (gfs2_is_ordered(ip))
1137 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1138 else if (gfs2_is_jdata(ip))
1139 inode->i_mapping->a_ops = &gfs2_jdata_aops;