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/backing-dev.h>
23 #include <linux/uio.h>
24 #include <trace/events/writeback.h>
41 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
42 unsigned int from, unsigned int len)
44 struct buffer_head *head = page_buffers(page);
45 unsigned int bsize = head->b_size;
46 struct buffer_head *bh;
47 unsigned int to = from + len;
48 unsigned int start, end;
50 for (bh = head, start = 0; bh != head || !start;
51 bh = bh->b_this_page, start = end) {
57 set_buffer_uptodate(bh);
58 gfs2_trans_add_data(ip->i_gl, bh);
63 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
65 * @lblock: The block number to look up
66 * @bh_result: The buffer head to return the result in
67 * @create: Non-zero if we may add block to the file
72 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
73 struct buffer_head *bh_result, int create)
77 error = gfs2_block_map(inode, lblock, bh_result, 0);
80 if (!buffer_mapped(bh_result))
85 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
86 struct buffer_head *bh_result, int create)
88 return gfs2_block_map(inode, lblock, bh_result, 0);
92 * gfs2_writepage_common - Common bits of writepage
93 * @page: The page to be written
94 * @wbc: The writeback control
96 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
99 static int gfs2_writepage_common(struct page *page,
100 struct writeback_control *wbc)
102 struct inode *inode = page->mapping->host;
103 struct gfs2_inode *ip = GFS2_I(inode);
104 struct gfs2_sbd *sdp = GFS2_SB(inode);
105 loff_t i_size = i_size_read(inode);
106 pgoff_t end_index = i_size >> PAGE_SHIFT;
109 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
111 if (current->journal_info)
113 /* Is the page fully outside i_size? (truncate in progress) */
114 offset = i_size & (PAGE_SIZE-1);
115 if (page->index > end_index || (page->index == end_index && !offset)) {
116 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
121 redirty_page_for_writepage(wbc, page);
128 * gfs2_writepage - Write page for writeback mappings
130 * @wbc: The writeback control
134 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
138 ret = gfs2_writepage_common(page, wbc);
142 return nobh_writepage(page, gfs2_get_block_noalloc, wbc);
145 /* This is the same as calling block_write_full_page, but it also
146 * writes pages outside of i_size
148 static int gfs2_write_full_page(struct page *page, get_block_t *get_block,
149 struct writeback_control *wbc)
151 struct inode * const inode = page->mapping->host;
152 loff_t i_size = i_size_read(inode);
153 const pgoff_t end_index = i_size >> PAGE_SHIFT;
157 * The page straddles i_size. It must be zeroed out on each and every
158 * writepage invocation because it may be mmapped. "A file is mapped
159 * in multiples of the page size. For a file that is not a multiple of
160 * the page size, the remaining memory is zeroed when mapped, and
161 * writes to that region are not written out to the file."
163 offset = i_size & (PAGE_SIZE-1);
164 if (page->index == end_index && offset)
165 zero_user_segment(page, offset, PAGE_SIZE);
167 return __block_write_full_page(inode, page, get_block, wbc,
168 end_buffer_async_write);
172 * __gfs2_jdata_writepage - The core of jdata writepage
173 * @page: The page to write
174 * @wbc: The writeback control
176 * This is shared between writepage and writepages and implements the
177 * core of the writepage operation. If a transaction is required then
178 * PageChecked will have been set and the transaction will have
179 * already been started before this is called.
182 static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
184 struct inode *inode = page->mapping->host;
185 struct gfs2_inode *ip = GFS2_I(inode);
186 struct gfs2_sbd *sdp = GFS2_SB(inode);
188 if (PageChecked(page)) {
189 ClearPageChecked(page);
190 if (!page_has_buffers(page)) {
191 create_empty_buffers(page, inode->i_sb->s_blocksize,
192 BIT(BH_Dirty)|BIT(BH_Uptodate));
194 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize);
196 return gfs2_write_full_page(page, gfs2_get_block_noalloc, wbc);
200 * gfs2_jdata_writepage - Write complete page
201 * @page: Page to write
202 * @wbc: The writeback control
208 static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc)
210 struct inode *inode = page->mapping->host;
211 struct gfs2_inode *ip = GFS2_I(inode);
212 struct gfs2_sbd *sdp = GFS2_SB(inode);
215 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl)))
217 if (PageChecked(page) || current->journal_info)
219 ret = __gfs2_jdata_writepage(page, wbc);
223 redirty_page_for_writepage(wbc, page);
230 * gfs2_writepages - Write a bunch of dirty pages back to disk
231 * @mapping: The mapping to write
232 * @wbc: Write-back control
234 * Used for both ordered and writeback modes.
236 static int gfs2_writepages(struct address_space *mapping,
237 struct writeback_control *wbc)
239 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
240 int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
243 * Even if we didn't write any pages here, we might still be holding
244 * dirty pages in the ail. We forcibly flush the ail because we don't
245 * want balance_dirty_pages() to loop indefinitely trying to write out
246 * pages held in the ail that it can't find.
249 set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags);
255 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
256 * @mapping: The mapping
257 * @wbc: The writeback control
258 * @pvec: The vector of pages
259 * @nr_pages: The number of pages to write
260 * @done_index: Page index
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,
271 struct inode *inode = mapping->host;
272 struct gfs2_sbd *sdp = GFS2_SB(inode);
273 unsigned nrblocks = nr_pages * (PAGE_SIZE/inode->i_sb->s_blocksize);
277 ret = gfs2_trans_begin(sdp, nrblocks, nrblocks);
281 for(i = 0; i < nr_pages; i++) {
282 struct page *page = pvec->pages[i];
284 *done_index = page->index;
288 if (unlikely(page->mapping != mapping)) {
294 if (!PageDirty(page)) {
295 /* someone wrote it for us */
296 goto continue_unlock;
299 if (PageWriteback(page)) {
300 if (wbc->sync_mode != WB_SYNC_NONE)
301 wait_on_page_writeback(page);
303 goto continue_unlock;
306 BUG_ON(PageWriteback(page));
307 if (!clear_page_dirty_for_io(page))
308 goto continue_unlock;
310 trace_wbc_writepage(wbc, inode_to_bdi(inode));
312 ret = __gfs2_jdata_writepage(page, wbc);
314 if (ret == AOP_WRITEPAGE_ACTIVATE) {
320 * done_index is set past this page,
321 * so media errors will not choke
322 * background writeout for the entire
323 * file. This has consequences for
324 * range_cyclic semantics (ie. it may
325 * not be suitable for data integrity
328 *done_index = page->index + 1;
335 * We stop writing back only if we are not doing
336 * integrity sync. In case of integrity sync we have to
337 * keep going until we have written all the pages
338 * we tagged for writeback prior to entering this loop.
340 if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) {
351 * gfs2_write_cache_jdata - Like write_cache_pages but different
352 * @mapping: The mapping to write
353 * @wbc: The writeback control
355 * The reason that we use our own function here is that we need to
356 * start transactions before we grab page locks. This allows us
357 * to get the ordering right.
360 static int gfs2_write_cache_jdata(struct address_space *mapping,
361 struct writeback_control *wbc)
367 pgoff_t uninitialized_var(writeback_index);
376 if (wbc->range_cyclic) {
377 writeback_index = mapping->writeback_index; /* prev offset */
378 index = writeback_index;
385 index = wbc->range_start >> PAGE_SHIFT;
386 end = wbc->range_end >> PAGE_SHIFT;
387 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
389 cycled = 1; /* ignore range_cyclic tests */
391 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
392 tag = PAGECACHE_TAG_TOWRITE;
394 tag = PAGECACHE_TAG_DIRTY;
397 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
398 tag_pages_for_writeback(mapping, index, end);
400 while (!done && (index <= end)) {
401 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
406 ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index);
411 pagevec_release(&pvec);
415 if (!cycled && !done) {
418 * We hit the last page and there is more work to be done: wrap
419 * back to the start of the file
423 end = writeback_index - 1;
427 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
428 mapping->writeback_index = done_index;
435 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
436 * @mapping: The mapping to write
437 * @wbc: The writeback control
441 static int gfs2_jdata_writepages(struct address_space *mapping,
442 struct writeback_control *wbc)
444 struct gfs2_inode *ip = GFS2_I(mapping->host);
445 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
448 ret = gfs2_write_cache_jdata(mapping, wbc);
449 if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) {
450 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
451 GFS2_LFC_JDATA_WPAGES);
452 ret = gfs2_write_cache_jdata(mapping, wbc);
458 * stuffed_readpage - Fill in a Linux page with stuffed file data
465 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
467 struct buffer_head *dibh;
468 u64 dsize = i_size_read(&ip->i_inode);
473 * Due to the order of unstuffing files and ->fault(), we can be
474 * asked for a zero page in the case of a stuffed file being extended,
475 * so we need to supply one here. It doesn't happen often.
477 if (unlikely(page->index)) {
478 zero_user(page, 0, PAGE_SIZE);
479 SetPageUptodate(page);
483 error = gfs2_meta_inode_buffer(ip, &dibh);
487 kaddr = kmap_atomic(page);
488 if (dsize > gfs2_max_stuffed_size(ip))
489 dsize = gfs2_max_stuffed_size(ip);
490 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
491 memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
492 kunmap_atomic(kaddr);
493 flush_dcache_page(page);
495 SetPageUptodate(page);
502 * __gfs2_readpage - readpage
503 * @file: The file to read a page for
504 * @page: The page to read
506 * This is the core of gfs2's readpage. It's used by the internal file
507 * reading code as in that case we already hold the glock. Also it's
508 * called by gfs2_readpage() once the required lock has been granted.
511 static int __gfs2_readpage(void *file, struct page *page)
513 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
514 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
517 if (gfs2_is_stuffed(ip)) {
518 error = stuffed_readpage(ip, page);
521 error = mpage_readpage(page, gfs2_block_map);
524 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
531 * gfs2_readpage - read a page of a file
532 * @file: The file to read
533 * @page: The page of the file
535 * This deals with the locking required. We have to unlock and
536 * relock the page in order to get the locking in the right
540 static int gfs2_readpage(struct file *file, struct page *page)
542 struct address_space *mapping = page->mapping;
543 struct gfs2_inode *ip = GFS2_I(mapping->host);
544 struct gfs2_holder gh;
548 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
549 error = gfs2_glock_nq(&gh);
552 error = AOP_TRUNCATED_PAGE;
554 if (page->mapping == mapping && !PageUptodate(page))
555 error = __gfs2_readpage(file, page);
560 gfs2_holder_uninit(&gh);
561 if (error && error != AOP_TRUNCATED_PAGE)
567 * gfs2_internal_read - read an internal file
568 * @ip: The gfs2 inode
569 * @buf: The buffer to fill
570 * @pos: The file position
571 * @size: The amount to read
575 int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos,
578 struct address_space *mapping = ip->i_inode.i_mapping;
579 unsigned long index = *pos / PAGE_SIZE;
580 unsigned offset = *pos & (PAGE_SIZE - 1);
588 if (offset + size > PAGE_SIZE)
589 amt = PAGE_SIZE - offset;
590 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
592 return PTR_ERR(page);
593 p = kmap_atomic(page);
594 memcpy(buf + copied, p + offset, amt);
600 } while(copied < size);
606 * gfs2_readpages - Read a bunch of pages at once
607 * @file: The file to read from
608 * @mapping: Address space info
609 * @pages: List of pages to read
610 * @nr_pages: Number of pages to read
613 * 1. This is only for readahead, so we can simply ignore any things
614 * which are slightly inconvenient (such as locking conflicts between
615 * the page lock and the glock) and return having done no I/O. Its
616 * obviously not something we'd want to do on too regular a basis.
617 * Any I/O we ignore at this time will be done via readpage later.
618 * 2. We don't handle stuffed files here we let readpage do the honours.
619 * 3. mpage_readpages() does most of the heavy lifting in the common case.
620 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
623 static int gfs2_readpages(struct file *file, struct address_space *mapping,
624 struct list_head *pages, unsigned nr_pages)
626 struct inode *inode = mapping->host;
627 struct gfs2_inode *ip = GFS2_I(inode);
628 struct gfs2_sbd *sdp = GFS2_SB(inode);
629 struct gfs2_holder gh;
632 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
633 ret = gfs2_glock_nq(&gh);
636 if (!gfs2_is_stuffed(ip))
637 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map);
640 gfs2_holder_uninit(&gh);
641 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
647 * gfs2_write_begin - Begin to write to a file
648 * @file: The file to write to
649 * @mapping: The mapping in which to write
650 * @pos: The file offset at which to start writing
651 * @len: Length of the write
652 * @flags: Various flags
653 * @pagep: Pointer to return the page
654 * @fsdata: Pointer to return fs data (unused by GFS2)
659 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
660 loff_t pos, unsigned len, unsigned flags,
661 struct page **pagep, void **fsdata)
663 struct gfs2_inode *ip = GFS2_I(mapping->host);
664 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
665 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
666 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
667 unsigned requested = 0;
670 pgoff_t index = pos >> PAGE_SHIFT;
671 unsigned from = pos & (PAGE_SIZE - 1);
674 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
675 error = gfs2_glock_nq(&ip->i_gh);
678 if (&ip->i_inode == sdp->sd_rindex) {
679 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
680 GL_NOCACHE, &m_ip->i_gh);
681 if (unlikely(error)) {
682 gfs2_glock_dq(&ip->i_gh);
687 alloc_required = gfs2_write_alloc_required(ip, pos, len);
689 if (alloc_required || gfs2_is_jdata(ip))
690 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
692 if (alloc_required) {
693 struct gfs2_alloc_parms ap = { .aflags = 0, };
694 requested = data_blocks + ind_blocks;
695 ap.target = requested;
696 error = gfs2_quota_lock_check(ip, &ap);
700 error = gfs2_inplace_reserve(ip, &ap);
705 rblocks = RES_DINODE + ind_blocks;
706 if (gfs2_is_jdata(ip))
707 rblocks += data_blocks ? data_blocks : 1;
708 if (ind_blocks || data_blocks)
709 rblocks += RES_STATFS + RES_QUOTA;
710 if (&ip->i_inode == sdp->sd_rindex)
711 rblocks += 2 * RES_STATFS;
713 rblocks += gfs2_rg_blocks(ip, requested);
715 error = gfs2_trans_begin(sdp, rblocks,
716 PAGE_SIZE/sdp->sd_sb.sb_bsize);
721 flags |= AOP_FLAG_NOFS;
722 page = grab_cache_page_write_begin(mapping, index, flags);
727 if (gfs2_is_stuffed(ip)) {
729 if (pos + len > gfs2_max_stuffed_size(ip)) {
730 error = gfs2_unstuff_dinode(ip, page);
733 } else if (!PageUptodate(page)) {
734 error = stuffed_readpage(ip, page);
740 error = __block_write_begin(page, from, len, gfs2_block_map);
749 if (alloc_required) {
750 gfs2_inplace_release(ip);
751 if (pos + len > ip->i_inode.i_size)
752 gfs2_trim_blocks(&ip->i_inode);
760 gfs2_inplace_release(ip);
763 gfs2_quota_unlock(ip);
765 if (&ip->i_inode == sdp->sd_rindex) {
766 gfs2_glock_dq(&m_ip->i_gh);
767 gfs2_holder_uninit(&m_ip->i_gh);
769 gfs2_glock_dq(&ip->i_gh);
771 gfs2_holder_uninit(&ip->i_gh);
776 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
777 * @inode: the rindex inode
779 static void adjust_fs_space(struct inode *inode)
781 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
782 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
783 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
784 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
785 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
786 struct buffer_head *m_bh, *l_bh;
787 u64 fs_total, new_free;
789 /* Total up the file system space, according to the latest rindex. */
790 fs_total = gfs2_ri_total(sdp);
791 if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0)
794 spin_lock(&sdp->sd_statfs_spin);
795 gfs2_statfs_change_in(m_sc, m_bh->b_data +
796 sizeof(struct gfs2_dinode));
797 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
798 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
801 spin_unlock(&sdp->sd_statfs_spin);
802 fs_warn(sdp, "File system extended by %llu blocks.\n",
803 (unsigned long long)new_free);
804 gfs2_statfs_change(sdp, new_free, new_free, 0);
806 if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0)
808 update_statfs(sdp, m_bh, l_bh);
815 * gfs2_stuffed_write_end - Write end for stuffed files
817 * @dibh: The buffer_head containing the on-disk inode
818 * @pos: The file position
819 * @copied: How much was actually copied by the VFS
822 * This copies the data from the page into the inode block after
823 * the inode data structure itself.
827 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
828 loff_t pos, unsigned copied,
831 struct gfs2_inode *ip = GFS2_I(inode);
832 u64 to = pos + copied;
834 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
836 BUG_ON(pos + copied > gfs2_max_stuffed_size(ip));
838 kaddr = kmap_atomic(page);
839 memcpy(buf + pos, kaddr + pos, copied);
840 flush_dcache_page(page);
841 kunmap_atomic(kaddr);
843 WARN_ON(!PageUptodate(page));
848 if (inode->i_size < to)
849 i_size_write(inode, to);
850 mark_inode_dirty(inode);
857 * @file: The file to write to
858 * @mapping: The address space to write to
859 * @pos: The file position
860 * @len: The length of the data
861 * @copied: How much was actually copied by the VFS
862 * @page: The page that has been written
863 * @fsdata: The fsdata (unused in GFS2)
865 * The main write_end function for GFS2. We just put our locking around the VFS
866 * provided functions.
871 static int gfs2_write_end(struct file *file, struct address_space *mapping,
872 loff_t pos, unsigned len, unsigned copied,
873 struct page *page, void *fsdata)
875 struct inode *inode = page->mapping->host;
876 struct gfs2_inode *ip = GFS2_I(inode);
877 struct gfs2_sbd *sdp = GFS2_SB(inode);
878 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
879 struct buffer_head *dibh;
881 struct gfs2_trans *tr = current->journal_info;
884 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
886 ret = gfs2_meta_inode_buffer(ip, &dibh);
890 if (gfs2_is_stuffed(ip)) {
891 ret = gfs2_stuffed_write_end(inode, dibh, pos, copied, page);
896 if (gfs2_is_jdata(ip))
897 gfs2_page_add_databufs(ip, page, pos & ~PAGE_MASK, len);
899 gfs2_ordered_add_inode(ip);
901 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
903 if (tr->tr_num_buf_new)
904 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
906 gfs2_trans_add_meta(ip->i_gl, dibh);
909 if (inode == sdp->sd_rindex) {
910 adjust_fs_space(inode);
911 sdp->sd_rindex_uptodate = 0;
921 gfs2_inplace_release(ip);
922 if (ip->i_qadata && ip->i_qadata->qa_qd_num)
923 gfs2_quota_unlock(ip);
924 if (inode == sdp->sd_rindex) {
925 gfs2_glock_dq(&m_ip->i_gh);
926 gfs2_holder_uninit(&m_ip->i_gh);
928 gfs2_glock_dq(&ip->i_gh);
929 gfs2_holder_uninit(&ip->i_gh);
934 * jdata_set_page_dirty - Page dirtying function
935 * @page: The page to dirty
937 * Returns: 1 if it dirtyed the page, or 0 otherwise
940 static int jdata_set_page_dirty(struct page *page)
942 SetPageChecked(page);
943 return __set_page_dirty_buffers(page);
947 * gfs2_bmap - Block map function
948 * @mapping: Address space info
949 * @lblock: The block to map
951 * Returns: The disk address for the block or 0 on hole or error
954 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
956 struct gfs2_inode *ip = GFS2_I(mapping->host);
957 struct gfs2_holder i_gh;
961 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
965 if (!gfs2_is_stuffed(ip))
966 dblock = generic_block_bmap(mapping, lblock, gfs2_block_map);
968 gfs2_glock_dq_uninit(&i_gh);
973 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
975 struct gfs2_bufdata *bd;
979 clear_buffer_dirty(bh);
982 if (!list_empty(&bd->bd_list) && !buffer_pinned(bh))
983 list_del_init(&bd->bd_list);
985 gfs2_remove_from_journal(bh, REMOVE_JDATA);
988 clear_buffer_mapped(bh);
989 clear_buffer_req(bh);
990 clear_buffer_new(bh);
991 gfs2_log_unlock(sdp);
995 static void gfs2_invalidatepage(struct page *page, unsigned int offset,
998 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
999 unsigned int stop = offset + length;
1000 int partial_page = (offset || length < PAGE_SIZE);
1001 struct buffer_head *bh, *head;
1002 unsigned long pos = 0;
1004 BUG_ON(!PageLocked(page));
1006 ClearPageChecked(page);
1007 if (!page_has_buffers(page))
1010 bh = head = page_buffers(page);
1012 if (pos + bh->b_size > stop)
1016 gfs2_discard(sdp, bh);
1018 bh = bh->b_this_page;
1019 } while (bh != head);
1022 try_to_release_page(page, 0);
1026 * gfs2_ok_for_dio - check that dio is valid on this file
1028 * @offset: The offset at which we are reading or writing
1030 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
1031 * 1 (to accept the i/o request)
1033 static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset)
1036 * Should we return an error here? I can't see that O_DIRECT for
1037 * a stuffed file makes any sense. For now we'll silently fall
1038 * back to buffered I/O
1040 if (gfs2_is_stuffed(ip))
1043 if (offset >= i_size_read(&ip->i_inode))
1050 static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
1052 struct file *file = iocb->ki_filp;
1053 struct inode *inode = file->f_mapping->host;
1054 struct address_space *mapping = inode->i_mapping;
1055 struct gfs2_inode *ip = GFS2_I(inode);
1056 loff_t offset = iocb->ki_pos;
1057 struct gfs2_holder gh;
1061 * Deferred lock, even if its a write, since we do no allocation
1062 * on this path. All we need change is atime, and this lock mode
1063 * ensures that other nodes have flushed their buffered read caches
1064 * (i.e. their page cache entries for this inode). We do not,
1065 * unfortunately have the option of only flushing a range like
1068 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
1069 rv = gfs2_glock_nq(&gh);
1072 rv = gfs2_ok_for_dio(ip, offset);
1074 goto out; /* dio not valid, fall back to buffered i/o */
1077 * Now since we are holding a deferred (CW) lock at this point, you
1078 * might be wondering why this is ever needed. There is a case however
1079 * where we've granted a deferred local lock against a cached exclusive
1080 * glock. That is ok provided all granted local locks are deferred, but
1081 * it also means that it is possible to encounter pages which are
1082 * cached and possibly also mapped. So here we check for that and sort
1083 * them out ahead of the dio. The glock state machine will take care of
1086 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
1087 * the first place, mapping->nr_pages will always be zero.
1089 if (mapping->nrpages) {
1090 loff_t lstart = offset & ~(PAGE_SIZE - 1);
1091 loff_t len = iov_iter_count(iter);
1092 loff_t end = PAGE_ALIGN(offset + len) - 1;
1097 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
1098 unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
1099 rv = filemap_write_and_wait_range(mapping, lstart, end);
1102 if (iov_iter_rw(iter) == WRITE)
1103 truncate_inode_pages_range(mapping, lstart, end);
1106 rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
1107 gfs2_get_block_direct, NULL, NULL, 0);
1111 gfs2_holder_uninit(&gh);
1116 * gfs2_releasepage - free the metadata associated with a page
1117 * @page: the page that's being released
1118 * @gfp_mask: passed from Linux VFS, ignored by us
1120 * Call try_to_free_buffers() if the buffers in this page can be
1126 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
1128 struct address_space *mapping = page->mapping;
1129 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
1130 struct buffer_head *bh, *head;
1131 struct gfs2_bufdata *bd;
1133 if (!page_has_buffers(page))
1137 * From xfs_vm_releasepage: mm accommodates an old ext3 case where
1138 * clean pages might not have had the dirty bit cleared. Thus, it can
1139 * send actual dirty pages to ->releasepage() via shrink_active_list().
1141 * As a workaround, we skip pages that contain dirty buffers below.
1142 * Once ->releasepage isn't called on dirty pages anymore, we can warn
1143 * on dirty buffers like we used to here again.
1147 spin_lock(&sdp->sd_ail_lock);
1148 head = bh = page_buffers(page);
1150 if (atomic_read(&bh->b_count))
1151 goto cannot_release;
1153 if (bd && bd->bd_tr)
1154 goto cannot_release;
1155 if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh)))
1156 goto cannot_release;
1157 bh = bh->b_this_page;
1158 } while(bh != head);
1159 spin_unlock(&sdp->sd_ail_lock);
1161 head = bh = page_buffers(page);
1165 gfs2_assert_warn(sdp, bd->bd_bh == bh);
1166 if (!list_empty(&bd->bd_list))
1167 list_del_init(&bd->bd_list);
1169 bh->b_private = NULL;
1170 kmem_cache_free(gfs2_bufdata_cachep, bd);
1173 bh = bh->b_this_page;
1174 } while (bh != head);
1175 gfs2_log_unlock(sdp);
1177 return try_to_free_buffers(page);
1180 spin_unlock(&sdp->sd_ail_lock);
1181 gfs2_log_unlock(sdp);
1185 static const struct address_space_operations gfs2_writeback_aops = {
1186 .writepage = gfs2_writepage,
1187 .writepages = gfs2_writepages,
1188 .readpage = gfs2_readpage,
1189 .readpages = gfs2_readpages,
1190 .write_begin = gfs2_write_begin,
1191 .write_end = gfs2_write_end,
1193 .invalidatepage = gfs2_invalidatepage,
1194 .releasepage = gfs2_releasepage,
1195 .direct_IO = gfs2_direct_IO,
1196 .migratepage = buffer_migrate_page,
1197 .is_partially_uptodate = block_is_partially_uptodate,
1198 .error_remove_page = generic_error_remove_page,
1201 static const struct address_space_operations gfs2_ordered_aops = {
1202 .writepage = gfs2_writepage,
1203 .writepages = gfs2_writepages,
1204 .readpage = gfs2_readpage,
1205 .readpages = gfs2_readpages,
1206 .write_begin = gfs2_write_begin,
1207 .write_end = gfs2_write_end,
1208 .set_page_dirty = __set_page_dirty_buffers,
1210 .invalidatepage = gfs2_invalidatepage,
1211 .releasepage = gfs2_releasepage,
1212 .direct_IO = gfs2_direct_IO,
1213 .migratepage = buffer_migrate_page,
1214 .is_partially_uptodate = block_is_partially_uptodate,
1215 .error_remove_page = generic_error_remove_page,
1218 static const struct address_space_operations gfs2_jdata_aops = {
1219 .writepage = gfs2_jdata_writepage,
1220 .writepages = gfs2_jdata_writepages,
1221 .readpage = gfs2_readpage,
1222 .readpages = gfs2_readpages,
1223 .write_begin = gfs2_write_begin,
1224 .write_end = gfs2_write_end,
1225 .set_page_dirty = jdata_set_page_dirty,
1227 .invalidatepage = gfs2_invalidatepage,
1228 .releasepage = gfs2_releasepage,
1229 .is_partially_uptodate = block_is_partially_uptodate,
1230 .error_remove_page = generic_error_remove_page,
1233 void gfs2_set_aops(struct inode *inode)
1235 struct gfs2_inode *ip = GFS2_I(inode);
1237 if (gfs2_is_writeback(ip))
1238 inode->i_mapping->a_ops = &gfs2_writeback_aops;
1239 else if (gfs2_is_ordered(ip))
1240 inode->i_mapping->a_ops = &gfs2_ordered_aops;
1241 else if (gfs2_is_jdata(ip))
1242 inode->i_mapping->a_ops = &gfs2_jdata_aops;