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>
32 #include "ops_address.h"
41 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
42 unsigned int from, unsigned int to)
44 struct buffer_head *head = page_buffers(page);
45 unsigned int bsize = head->b_size;
46 struct buffer_head *bh;
47 unsigned int start, end;
49 for (bh = head, start = 0; bh != head || !start;
50 bh = bh->b_this_page, start = end) {
52 if (end <= from || start >= to)
54 if (gfs2_is_jdata(ip))
55 set_buffer_uptodate(bh);
56 gfs2_trans_add_bh(ip->i_gl, bh, 0);
61 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
63 * @lblock: The block number to look up
64 * @bh_result: The buffer head to return the result in
65 * @create: Non-zero if we may add block to the file
70 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
71 struct buffer_head *bh_result, int create)
75 error = gfs2_block_map(inode, lblock, bh_result, 0);
78 if (!buffer_mapped(bh_result))
83 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
84 struct buffer_head *bh_result, int create)
86 return gfs2_block_map(inode, lblock, bh_result, 0);
90 * gfs2_writepage_common - Common bits of writepage
91 * @page: The page to be written
92 * @wbc: The writeback control
94 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
97 static int gfs2_writepage_common(struct page *page,
98 struct writeback_control *wbc)
100 struct inode *inode = page->mapping->host;
101 struct gfs2_inode *ip = GFS2_I(inode);
102 struct gfs2_sbd *sdp = GFS2_SB(inode);
103 loff_t i_size = i_size_read(inode);
104 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
107 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
109 if (current->journal_info)
111 /* Is the page fully outside i_size? (truncate in progress) */
112 offset = i_size & (PAGE_CACHE_SIZE-1);
113 if (page->index > end_index || (page->index == end_index && !offset)) {
114 page->mapping->a_ops->invalidatepage(page, 0);
119 redirty_page_for_writepage(wbc, page);
126 * gfs2_writeback_writepage - Write page for writeback mappings
128 * @wbc: The writeback control
132 static int gfs2_writeback_writepage(struct page *page,
133 struct writeback_control *wbc)
137 ret = gfs2_writepage_common(page, wbc);
141 ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc);
143 ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
148 * gfs2_ordered_writepage - Write page for ordered data files
149 * @page: The page to write
150 * @wbc: The writeback control
154 static int gfs2_ordered_writepage(struct page *page,
155 struct writeback_control *wbc)
157 struct inode *inode = page->mapping->host;
158 struct gfs2_inode *ip = GFS2_I(inode);
161 ret = gfs2_writepage_common(page, wbc);
165 if (!page_has_buffers(page)) {
166 create_empty_buffers(page, inode->i_sb->s_blocksize,
167 (1 << BH_Dirty)|(1 << BH_Uptodate));
169 gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1);
170 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
174 * __gfs2_jdata_writepage - The core of jdata writepage
175 * @page: The page to write
176 * @wbc: The writeback control
178 * This is shared between writepage and writepages and implements the
179 * core of the writepage operation. If a transaction is required then
180 * PageChecked will have been set and the transaction will have
181 * already been started before this is called.
184 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
186 struct inode *inode = page->mapping->host;
187 struct gfs2_inode *ip = GFS2_I(inode);
188 struct gfs2_sbd *sdp = GFS2_SB(inode);
190 if (PageChecked(page)) {
191 ClearPageChecked(page);
192 if (!page_has_buffers(page)) {
193 create_empty_buffers(page, inode->i_sb->s_blocksize,
194 (1 << BH_Dirty)|(1 << BH_Uptodate));
196 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
198 return block_write_full_page(page, gfs2_get_block_noalloc, wbc);
202 * gfs2_jdata_writepage - Write complete page
203 * @page: Page to write
209 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
211 struct inode *inode = page->mapping->host;
212 struct gfs2_sbd *sdp = GFS2_SB(inode);
216 error = gfs2_writepage_common(page, wbc);
220 if (PageChecked(page)) {
221 if (wbc->sync_mode != WB_SYNC_ALL)
223 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
228 error = __gfs2_jdata_writepage(page, wbc);
234 redirty_page_for_writepage(wbc, page);
240 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
241 * @mapping: The mapping to write
242 * @wbc: Write-back control
244 * For the data=writeback case we can already ignore buffer heads
245 * and write whole extents at once. This is a big reduction in the
246 * number of I/O requests we send and the bmap calls we make in this case.
248 static int gfs2_writeback_writepages(struct address_space *mapping,
249 struct writeback_control *wbc)
251 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
255 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
256 * @mapping: The mapping
257 * @wbc: The writeback control
258 * @writepage: The writepage function to call for each page
259 * @pvec: The vector of pages
260 * @nr_pages: The number of pages to write
262 * Returns: non-zero if loop should terminate, zero otherwise
265 static int gfs2_write_jdata_pagevec(struct address_space *mapping,
266 struct writeback_control *wbc,
267 struct pagevec *pvec,
268 int nr_pages, pgoff_t end)
270 struct inode *inode = mapping->host;
271 struct gfs2_sbd *sdp = GFS2_SB(inode);
272 loff_t i_size = i_size_read(inode);
273 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
274 unsigned offset = i_size & (PAGE_CACHE_SIZE-1);
275 unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize);
276 struct backing_dev_info *bdi = mapping->backing_dev_info;
280 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
284 for(i = 0; i < nr_pages; i++) {
285 struct page *page = pvec->pages[i];
289 if (unlikely(page->mapping != mapping)) {
294 if (!wbc->range_cyclic && page->index > end) {
300 if (wbc->sync_mode != WB_SYNC_NONE)
301 wait_on_page_writeback(page);
303 if (PageWriteback(page) ||
304 !clear_page_dirty_for_io(page)) {
309 /* Is the page fully outside i_size? (truncate in progress) */
310 if (page->index > end_index || (page->index == end_index && !offset)) {
311 page->mapping->a_ops->invalidatepage(page, 0);
316 ret = __gfs2_jdata_writepage(page, wbc);
318 if (ret || (--(wbc->nr_to_write) <= 0))
320 if (wbc->nonblocking && bdi_write_congested(bdi)) {
321 wbc->encountered_congestion = 1;
331 * gfs2_write_cache_jdata - Like write_cache_pages but different
332 * @mapping: The mapping to write
333 * @wbc: The writeback control
334 * @writepage: The writepage function to call
335 * @data: The data to pass to writepage
337 * The reason that we use our own function here is that we need to
338 * start transactions before we grab page locks. This allows us
339 * to get the ordering right.
342 static int gfs2_write_cache_jdata(struct address_space *mapping,
343 struct writeback_control *wbc)
345 struct backing_dev_info *bdi = mapping->backing_dev_info;
355 if (wbc->nonblocking && bdi_write_congested(bdi)) {
356 wbc->encountered_congestion = 1;
360 pagevec_init(&pvec, 0);
361 if (wbc->range_cyclic) {
362 index = mapping->writeback_index; /* Start from prev offset */
365 index = wbc->range_start >> PAGE_CACHE_SHIFT;
366 end = wbc->range_end >> PAGE_CACHE_SHIFT;
367 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
373 while (!done && (index <= end) &&
374 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
376 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) {
378 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end);
384 pagevec_release(&pvec);
388 if (!scanned && !done) {
390 * We hit the last page and there is more work to be done: wrap
391 * back to the start of the file
398 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
399 mapping->writeback_index = index;
405 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
406 * @mapping: The mapping to write
407 * @wbc: The writeback control
411 static int gfs2_jdata_writepages(struct address_space *mapping,
412 struct writeback_control *wbc)
414 struct gfs2_inode *ip = GFS2_I(mapping->host);
415 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
418 ret = gfs2_write_cache_jdata(mapping, wbc);
419 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
420 gfs2_log_flush(sdp, ip->i_gl);
421 ret = gfs2_write_cache_jdata(mapping, wbc);
427 * stuffed_readpage - Fill in a Linux page with stuffed file data
434 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
436 struct buffer_head *dibh;
441 * Due to the order of unstuffing files and ->nopage(), we can be
442 * asked for a zero page in the case of a stuffed file being extended,
443 * so we need to supply one here. It doesn't happen often.
445 if (unlikely(page->index)) {
446 zero_user(page, 0, PAGE_CACHE_SIZE);
450 error = gfs2_meta_inode_buffer(ip, &dibh);
454 kaddr = kmap_atomic(page, KM_USER0);
455 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
457 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
458 kunmap_atomic(kaddr, KM_USER0);
459 flush_dcache_page(page);
461 SetPageUptodate(page);
468 * __gfs2_readpage - readpage
469 * @file: The file to read a page for
470 * @page: The page to read
472 * This is the core of gfs2's readpage. Its used by the internal file
473 * reading code as in that case we already hold the glock. Also its
474 * called by gfs2_readpage() once the required lock has been granted.
478 static int __gfs2_readpage(void *file, struct page *page)
480 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
481 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
484 if (gfs2_is_stuffed(ip)) {
485 error = stuffed_readpage(ip, page);
488 error = mpage_readpage(page, gfs2_block_map);
491 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
498 * gfs2_readpage - read a page of a file
499 * @file: The file to read
500 * @page: The page of the file
502 * This deals with the locking required. We use a trylock in order to
503 * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
504 * in the event that we are unable to get the lock.
507 static int gfs2_readpage(struct file *file, struct page *page)
509 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
510 struct gfs2_holder *gh;
513 gh = gfs2_glock_is_locked_by_me(ip->i_gl);
515 gh = kmalloc(sizeof(struct gfs2_holder), GFP_NOFS);
518 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, gh);
520 error = gfs2_glock_nq_atime(gh);
521 if (likely(error != 0))
523 return AOP_TRUNCATED_PAGE;
525 error = __gfs2_readpage(file, page);
528 gfs2_holder_uninit(gh);
534 * gfs2_internal_read - read an internal file
535 * @ip: The gfs2 inode
536 * @ra_state: The readahead state (or NULL for no readahead)
537 * @buf: The buffer to fill
538 * @pos: The file position
539 * @size: The amount to read
543 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
544 char *buf, loff_t *pos, unsigned size)
546 struct address_space *mapping = ip->i_inode.i_mapping;
547 unsigned long index = *pos / PAGE_CACHE_SIZE;
548 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
556 if (offset + size > PAGE_CACHE_SIZE)
557 amt = PAGE_CACHE_SIZE - offset;
558 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
560 return PTR_ERR(page);
561 p = kmap_atomic(page, KM_USER0);
562 memcpy(buf + copied, p + offset, amt);
563 kunmap_atomic(p, KM_USER0);
564 mark_page_accessed(page);
565 page_cache_release(page);
569 } while(copied < size);
575 * gfs2_readpages - Read a bunch of pages at once
578 * 1. This is only for readahead, so we can simply ignore any things
579 * which are slightly inconvenient (such as locking conflicts between
580 * the page lock and the glock) and return having done no I/O. Its
581 * obviously not something we'd want to do on too regular a basis.
582 * Any I/O we ignore at this time will be done via readpage later.
583 * 2. We don't handle stuffed files here we let readpage do the honours.
584 * 3. mpage_readpages() does most of the heavy lifting in the common case.
585 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
588 static int gfs2_readpages(struct file *file, struct address_space *mapping,
589 struct list_head *pages, unsigned nr_pages)
591 struct inode *inode = mapping->host;
592 struct gfs2_inode *ip = GFS2_I(inode);
593 struct gfs2_sbd *sdp = GFS2_SB(inode);
594 struct gfs2_holder gh;
597 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
598 ret = gfs2_glock_nq_atime(&gh);
601 if (!gfs2_is_stuffed(ip))
602 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
605 gfs2_holder_uninit(&gh);
606 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
612 * gfs2_write_begin - Begin to write to a file
613 * @file: The file to write to
614 * @mapping: The mapping in which to write
615 * @pos: The file offset at which to start writing
616 * @len: Length of the write
617 * @flags: Various flags
618 * @pagep: Pointer to return the page
619 * @fsdata: Pointer to return fs data (unused by GFS2)
624 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
625 loff_t pos, unsigned len, unsigned flags,
626 struct page **pagep, void **fsdata)
628 struct gfs2_inode *ip = GFS2_I(mapping->host);
629 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
630 unsigned int data_blocks, ind_blocks, rblocks;
633 struct gfs2_alloc *al;
634 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
635 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
636 unsigned to = from + len;
639 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
640 error = gfs2_glock_nq_atime(&ip->i_gh);
644 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
645 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
649 if (alloc_required) {
650 al = gfs2_alloc_get(ip);
656 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
660 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
664 al->al_requested = data_blocks + ind_blocks;
665 error = gfs2_inplace_reserve(ip);
670 rblocks = RES_DINODE + ind_blocks;
671 if (gfs2_is_jdata(ip))
672 rblocks += data_blocks ? data_blocks : 1;
673 if (ind_blocks || data_blocks)
674 rblocks += RES_STATFS + RES_QUOTA;
676 error = gfs2_trans_begin(sdp, rblocks,
677 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
682 page = __grab_cache_page(mapping, index);
687 if (gfs2_is_stuffed(ip)) {
689 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
690 error = gfs2_unstuff_dinode(ip, page);
693 } else if (!PageUptodate(page)) {
694 error = stuffed_readpage(ip, page);
700 error = block_prepare_write(page, from, to, gfs2_block_map);
705 page_cache_release(page);
706 if (pos + len > ip->i_inode.i_size)
707 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
711 if (alloc_required) {
712 gfs2_inplace_release(ip);
714 gfs2_quota_unlock(ip);
719 gfs2_glock_dq(&ip->i_gh);
721 gfs2_holder_uninit(&ip->i_gh);
726 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
727 * @inode: the rindex inode
729 static void adjust_fs_space(struct inode *inode)
731 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
732 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
733 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
734 u64 fs_total, new_free;
736 /* Total up the file system space, according to the latest rindex. */
737 fs_total = gfs2_ri_total(sdp);
739 spin_lock(&sdp->sd_statfs_spin);
740 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
741 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
744 spin_unlock(&sdp->sd_statfs_spin);
745 fs_warn(sdp, "File system extended by %llu blocks.\n",
746 (unsigned long long)new_free);
747 gfs2_statfs_change(sdp, new_free, new_free, 0);
751 * gfs2_stuffed_write_end - Write end for stuffed files
753 * @dibh: The buffer_head containing the on-disk inode
754 * @pos: The file position
755 * @len: The length of the write
756 * @copied: How much was actually copied by the VFS
759 * This copies the data from the page into the inode block after
760 * the inode data structure itself.
764 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
765 loff_t pos, unsigned len, unsigned copied,
768 struct gfs2_inode *ip = GFS2_I(inode);
769 struct gfs2_sbd *sdp = GFS2_SB(inode);
770 u64 to = pos + copied;
772 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
773 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
775 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
776 kaddr = kmap_atomic(page, KM_USER0);
777 memcpy(buf + pos, kaddr + pos, copied);
778 memset(kaddr + pos + copied, 0, len - copied);
779 flush_dcache_page(page);
780 kunmap_atomic(kaddr, KM_USER0);
782 if (!PageUptodate(page))
783 SetPageUptodate(page);
785 page_cache_release(page);
787 if (inode->i_size < to) {
788 i_size_write(inode, to);
789 ip->i_di.di_size = inode->i_size;
790 di->di_size = cpu_to_be64(inode->i_size);
791 mark_inode_dirty(inode);
794 if (inode == sdp->sd_rindex)
795 adjust_fs_space(inode);
799 gfs2_glock_dq(&ip->i_gh);
800 gfs2_holder_uninit(&ip->i_gh);
806 * @file: The file to write to
807 * @mapping: The address space to write to
808 * @pos: The file position
809 * @len: The length of the data
811 * @page: The page that has been written
812 * @fsdata: The fsdata (unused in GFS2)
814 * The main write_end function for GFS2. We have a separate one for
815 * stuffed files as they are slightly different, otherwise we just
816 * put our locking around the VFS provided functions.
821 static int gfs2_write_end(struct file *file, struct address_space *mapping,
822 loff_t pos, unsigned len, unsigned copied,
823 struct page *page, void *fsdata)
825 struct inode *inode = page->mapping->host;
826 struct gfs2_inode *ip = GFS2_I(inode);
827 struct gfs2_sbd *sdp = GFS2_SB(inode);
828 struct buffer_head *dibh;
829 struct gfs2_alloc *al = ip->i_alloc;
830 struct gfs2_dinode *di;
831 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
832 unsigned int to = from + len;
835 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
837 ret = gfs2_meta_inode_buffer(ip, &dibh);
840 page_cache_release(page);
844 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
846 if (gfs2_is_stuffed(ip))
847 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
849 if (!gfs2_is_writeback(ip))
850 gfs2_page_add_databufs(ip, page, from, to);
852 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
854 if (likely(ret >= 0) && (inode->i_size > ip->i_di.di_size)) {
855 di = (struct gfs2_dinode *)dibh->b_data;
856 ip->i_di.di_size = inode->i_size;
857 di->di_size = cpu_to_be64(inode->i_size);
858 mark_inode_dirty(inode);
861 if (inode == sdp->sd_rindex)
862 adjust_fs_space(inode);
868 gfs2_inplace_release(ip);
869 gfs2_quota_unlock(ip);
872 gfs2_glock_dq(&ip->i_gh);
873 gfs2_holder_uninit(&ip->i_gh);
878 * gfs2_set_page_dirty - Page dirtying function
879 * @page: The page to dirty
881 * Returns: 1 if it dirtyed the page, or 0 otherwise
884 static int gfs2_set_page_dirty(struct page *page)
886 SetPageChecked(page);
887 return __set_page_dirty_buffers(page);
891 * gfs2_bmap - Block map function
892 * @mapping: Address space info
893 * @lblock: The block to map
895 * Returns: The disk address for the block or 0 on hole or error
898 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
900 struct gfs2_inode *ip = GFS2_I(mapping->host);
901 struct gfs2_holder i_gh;
905 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
909 if (!gfs2_is_stuffed(ip))
910 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
912 gfs2_glock_dq_uninit(&i_gh);
917 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
919 struct gfs2_bufdata *bd;
923 clear_buffer_dirty(bh);
926 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
927 list_del_init(&bd->bd_le.le_list);
929 gfs2_remove_from_journal(bh, current->journal_info, 0);
932 clear_buffer_mapped(bh);
933 clear_buffer_req(bh);
934 clear_buffer_new(bh);
935 gfs2_log_unlock(sdp);
939 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
941 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
942 struct buffer_head *bh, *head;
943 unsigned long pos = 0;
945 BUG_ON(!PageLocked(page));
947 ClearPageChecked(page);
948 if (!page_has_buffers(page))
951 bh = head = page_buffers(page);
954 gfs2_discard(sdp, bh);
956 bh = bh->b_this_page;
957 } while (bh != head);
960 try_to_release_page(page, 0);
964 * gfs2_ok_for_dio - check that dio is valid on this file
967 * @offset: The offset at which we are reading or writing
969 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
970 * 1 (to accept the i/o request)
972 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
975 * Should we return an error here? I can't see that O_DIRECT for
976 * a stuffed file makes any sense. For now we'll silently fall
977 * back to buffered I/O
979 if (gfs2_is_stuffed(ip))
982 if (offset > i_size_read(&ip->i_inode))
989 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
990 const struct iovec *iov, loff_t offset,
991 unsigned long nr_segs)
993 struct file *file = iocb->ki_filp;
994 struct inode *inode = file->f_mapping->host;
995 struct gfs2_inode *ip = GFS2_I(inode);
996 struct gfs2_holder gh;
1000 * Deferred lock, even if its a write, since we do no allocation
1001 * on this path. All we need change is atime, and this lock mode
1002 * ensures that other nodes have flushed their buffered read caches
1003 * (i.e. their page cache entries for this inode). We do not,
1004 * unfortunately have the option of only flushing a range like
1007 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
1008 rv = gfs2_glock_nq_atime(&gh);
1011 rv = gfs2_ok_for_dio(ip, rw, offset);
1013 goto out; /* dio not valid, fall back to buffered i/o */
1015 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
1016 iov, offset, nr_segs,
1017 gfs2_get_block_direct, NULL);
1019 gfs2_glock_dq_m(1, &gh);
1020 gfs2_holder_uninit(&gh);
1025 * gfs2_releasepage - free the metadata associated with a page
1026 * @page: the page that's being released
1027 * @gfp_mask: passed from Linux VFS, ignored by us
1029 * Call try_to_free_buffers() if the buffers in this page can be
1035 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1037 struct inode *aspace = page->mapping->host;
1038 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
1039 struct buffer_head *bh, *head;
1040 struct gfs2_bufdata *bd;
1042 if (!page_has_buffers(page))
1046 head = bh = page_buffers(page);
1048 if (atomic_read(&bh->b_count))
1049 goto cannot_release;
1051 if (bd && bd->bd_ail)
1052 goto cannot_release;
1053 gfs2_assert_warn(sdp, !buffer_pinned(bh));
1054 gfs2_assert_warn(sdp, !buffer_dirty(bh));
1055 bh = bh->b_this_page;
1056 } while(bh != head);
1057 gfs2_log_unlock(sdp);
1059 head = bh = page_buffers(page);
1064 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1065 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
1066 if (!list_empty(&bd->bd_le.le_list)) {
1067 if (!buffer_pinned(bh))
1068 list_del_init(&bd->bd_le.le_list);
1074 bh->b_private = NULL;
1076 gfs2_log_unlock(sdp);
1078 kmem_cache_free(gfs2_bufdata_cachep, bd);
1080 bh = bh->b_this_page;
1081 } while (bh != head);
1083 return try_to_free_buffers(page);
1085 gfs2_log_unlock(sdp);
1089 static const struct address_space_operations gfs2_writeback_aops = {
1090 .writepage = gfs2_writeback_writepage,
1091 .writepages = gfs2_writeback_writepages,
1092 .readpage = gfs2_readpage,
1093 .readpages = gfs2_readpages,
1094 .sync_page = block_sync_page,
1095 .write_begin = gfs2_write_begin,
1096 .write_end = gfs2_write_end,
1098 .invalidatepage = gfs2_invalidatepage,
1099 .releasepage = gfs2_releasepage,
1100 .direct_IO = gfs2_direct_IO,
1101 .migratepage = buffer_migrate_page,
1104 static const struct address_space_operations gfs2_ordered_aops = {
1105 .writepage = gfs2_ordered_writepage,
1106 .readpage = gfs2_readpage,
1107 .readpages = gfs2_readpages,
1108 .sync_page = block_sync_page,
1109 .write_begin = gfs2_write_begin,
1110 .write_end = gfs2_write_end,
1111 .set_page_dirty = gfs2_set_page_dirty,
1113 .invalidatepage = gfs2_invalidatepage,
1114 .releasepage = gfs2_releasepage,
1115 .direct_IO = gfs2_direct_IO,
1116 .migratepage = buffer_migrate_page,
1119 static const struct address_space_operations gfs2_jdata_aops = {
1120 .writepage = gfs2_jdata_writepage,
1121 .writepages = gfs2_jdata_writepages,
1122 .readpage = gfs2_readpage,
1123 .readpages = gfs2_readpages,
1124 .sync_page = block_sync_page,
1125 .write_begin = gfs2_write_begin,
1126 .write_end = gfs2_write_end,
1127 .set_page_dirty = gfs2_set_page_dirty,
1129 .invalidatepage = gfs2_invalidatepage,
1130 .releasepage = gfs2_releasepage,
1133 void gfs2_set_aops(struct inode *inode)
1135 struct gfs2_inode *ip = GFS2_I(inode);
1137 if (gfs2_is_writeback(ip))
1138 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1139 else if (gfs2_is_ordered(ip))
1140 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1141 else if (gfs2_is_jdata(ip))
1142 inode->i_mapping->a_ops = &gfs2_jdata_aops;