2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 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/gfs2_ondisk.h>
21 #include <linux/lm_interface.h>
30 #include "ops_address.h"
40 static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
41 unsigned int from, unsigned int to)
43 struct buffer_head *head = page_buffers(page);
44 unsigned int bsize = head->b_size;
45 struct buffer_head *bh;
46 unsigned int start, end;
48 for (bh = head, start = 0; bh != head || !start;
49 bh = bh->b_this_page, start = end) {
51 if (end <= from || start >= to)
53 if (gfs2_is_jdata(ip))
54 set_buffer_uptodate(bh);
55 gfs2_trans_add_bh(ip->i_gl, bh, 0);
60 * gfs2_get_block - Fills in a buffer head with details about a block
62 * @lblock: The block number to look up
63 * @bh_result: The buffer head to return the result in
64 * @create: Non-zero if we may add block to the file
69 int gfs2_get_block(struct inode *inode, sector_t lblock,
70 struct buffer_head *bh_result, int create)
72 return gfs2_block_map(inode, lblock, create, bh_result);
76 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
78 * @lblock: The block number to look up
79 * @bh_result: The buffer head to return the result in
80 * @create: Non-zero if we may add block to the file
85 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
86 struct buffer_head *bh_result, int create)
90 error = gfs2_block_map(inode, lblock, 0, bh_result);
93 if (bh_result->b_blocknr == 0)
98 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
99 struct buffer_head *bh_result, int create)
101 return gfs2_block_map(inode, lblock, 0, bh_result);
105 * gfs2_writepage - Write complete page
106 * @page: Page to write
110 * Some of this is copied from block_write_full_page() although we still
111 * call it to do most of the work.
114 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
116 struct inode *inode = page->mapping->host;
117 struct gfs2_inode *ip = GFS2_I(inode);
118 struct gfs2_sbd *sdp = GFS2_SB(inode);
119 loff_t i_size = i_size_read(inode);
120 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
125 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
129 if (current->journal_info)
132 /* Is the page fully outside i_size? (truncate in progress) */
133 offset = i_size & (PAGE_CACHE_SIZE-1);
134 if (page->index > end_index || (page->index == end_index && !offset)) {
135 page->mapping->a_ops->invalidatepage(page, 0);
137 return 0; /* don't care */
140 if ((sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) &&
142 ClearPageChecked(page);
143 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
146 if (!page_has_buffers(page)) {
147 create_empty_buffers(page, inode->i_sb->s_blocksize,
148 (1 << BH_Dirty)|(1 << BH_Uptodate));
150 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
153 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
156 gfs2_meta_cache_flush(ip);
160 redirty_page_for_writepage(wbc, page);
166 * gfs2_writepages - Write a bunch of dirty pages back to disk
167 * @mapping: The mapping to write
168 * @wbc: Write-back control
170 * For journaled files and/or ordered writes this just falls back to the
171 * kernel's default writepages path for now. We will probably want to change
172 * that eventually (i.e. when we look at allocate on flush).
174 * For the data=writeback case though we can already ignore buffer heads
175 * and write whole extents at once. This is a big reduction in the
176 * number of I/O requests we send and the bmap calls we make in this case.
178 static int gfs2_writepages(struct address_space *mapping,
179 struct writeback_control *wbc)
181 struct inode *inode = mapping->host;
182 struct gfs2_inode *ip = GFS2_I(inode);
183 struct gfs2_sbd *sdp = GFS2_SB(inode);
185 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
186 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
188 return generic_writepages(mapping, wbc);
192 * stuffed_readpage - Fill in a Linux page with stuffed file data
199 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
201 struct buffer_head *dibh;
206 * Due to the order of unstuffing files and ->nopage(), we can be
207 * asked for a zero page in the case of a stuffed file being extended,
208 * so we need to supply one here. It doesn't happen often.
210 if (unlikely(page->index)) {
211 zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
215 error = gfs2_meta_inode_buffer(ip, &dibh);
219 kaddr = kmap_atomic(page, KM_USER0);
220 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
222 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
223 kunmap_atomic(kaddr, KM_USER0);
224 flush_dcache_page(page);
226 SetPageUptodate(page);
233 * gfs2_readpage - readpage with locking
234 * @file: The file to read a page for. N.B. This may be NULL if we are
235 * reading an internal file.
236 * @page: The page to read
241 static int gfs2_readpage(struct file *file, struct page *page)
243 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
244 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
245 struct gfs2_file *gf = NULL;
246 struct gfs2_holder gh;
250 if (likely(file != &gfs2_internal_file_sentinel)) {
252 gf = file->private_data;
253 if (test_bit(GFF_EXLOCK, &gf->f_flags))
254 /* gfs2_sharewrite_fault has grabbed the ip->i_gl already */
257 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
259 error = gfs2_glock_nq_atime(&gh);
265 if (gfs2_is_stuffed(ip)) {
266 error = stuffed_readpage(ip, page);
269 error = mpage_readpage(page, gfs2_get_block);
271 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
275 gfs2_glock_dq_m(1, &gh);
276 gfs2_holder_uninit(&gh);
282 if (error == GLR_TRYFAILED) {
283 error = AOP_TRUNCATED_PAGE;
287 gfs2_holder_uninit(&gh);
292 * gfs2_readpages - Read a bunch of pages at once
295 * 1. This is only for readahead, so we can simply ignore any things
296 * which are slightly inconvenient (such as locking conflicts between
297 * the page lock and the glock) and return having done no I/O. Its
298 * obviously not something we'd want to do on too regular a basis.
299 * Any I/O we ignore at this time will be done via readpage later.
300 * 2. We don't handle stuffed files here we let readpage do the honours.
301 * 3. mpage_readpages() does most of the heavy lifting in the common case.
302 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
303 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
304 * well as read-ahead.
306 static int gfs2_readpages(struct file *file, struct address_space *mapping,
307 struct list_head *pages, unsigned nr_pages)
309 struct inode *inode = mapping->host;
310 struct gfs2_inode *ip = GFS2_I(inode);
311 struct gfs2_sbd *sdp = GFS2_SB(inode);
312 struct gfs2_holder gh;
316 if (likely(file != &gfs2_internal_file_sentinel)) {
318 struct gfs2_file *gf = file->private_data;
319 if (test_bit(GFF_EXLOCK, &gf->f_flags))
322 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
323 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
325 ret = gfs2_glock_nq_atime(&gh);
326 if (ret == GLR_TRYFAILED)
332 if (!gfs2_is_stuffed(ip))
333 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
336 gfs2_glock_dq_m(1, &gh);
337 gfs2_holder_uninit(&gh);
340 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
347 gfs2_holder_uninit(&gh);
352 * gfs2_prepare_write - Prepare to write a page to a file
353 * @file: The file to write to
354 * @page: The page which is to be prepared for writing
355 * @from: From (byte range within page)
356 * @to: To (byte range within page)
361 static int gfs2_prepare_write(struct file *file, struct page *page,
362 unsigned from, unsigned to)
364 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
365 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
366 unsigned int data_blocks, ind_blocks, rblocks;
369 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
370 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
371 struct gfs2_alloc *al;
372 unsigned int write_len = to - from;
375 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
376 error = gfs2_glock_nq_atime(&ip->i_gh);
377 if (unlikely(error)) {
378 if (error == GLR_TRYFAILED) {
380 error = AOP_TRUNCATED_PAGE;
386 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
388 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
393 ip->i_alloc.al_requested = 0;
394 if (alloc_required) {
395 al = gfs2_alloc_get(ip);
397 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
401 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
405 al->al_requested = data_blocks + ind_blocks;
406 error = gfs2_inplace_reserve(ip);
411 rblocks = RES_DINODE + ind_blocks;
412 if (gfs2_is_jdata(ip))
413 rblocks += data_blocks ? data_blocks : 1;
414 if (ind_blocks || data_blocks)
415 rblocks += RES_STATFS + RES_QUOTA;
417 error = gfs2_trans_begin(sdp, rblocks, 0);
421 if (gfs2_is_stuffed(ip)) {
422 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
423 error = gfs2_unstuff_dinode(ip, page);
426 } else if (!PageUptodate(page))
427 error = stuffed_readpage(ip, page);
432 error = block_prepare_write(page, from, to, gfs2_get_block);
438 if (alloc_required) {
439 gfs2_inplace_release(ip);
441 gfs2_quota_unlock(ip);
446 gfs2_glock_dq_m(1, &ip->i_gh);
448 gfs2_holder_uninit(&ip->i_gh);
455 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
456 * @inode: the rindex inode
458 static void adjust_fs_space(struct inode *inode)
460 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
461 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
462 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
463 u64 fs_total, new_free;
465 /* Total up the file system space, according to the latest rindex. */
466 fs_total = gfs2_ri_total(sdp);
468 spin_lock(&sdp->sd_statfs_spin);
469 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
470 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
473 spin_unlock(&sdp->sd_statfs_spin);
474 fs_warn(sdp, "File system extended by %llu blocks.\n",
475 (unsigned long long)new_free);
476 gfs2_statfs_change(sdp, new_free, new_free, 0);
480 * gfs2_commit_write - Commit write to a file
481 * @file: The file to write to
482 * @page: The page containing the data
483 * @from: From (byte range within page)
484 * @to: To (byte range within page)
489 static int gfs2_commit_write(struct file *file, struct page *page,
490 unsigned from, unsigned to)
492 struct inode *inode = page->mapping->host;
493 struct gfs2_inode *ip = GFS2_I(inode);
494 struct gfs2_sbd *sdp = GFS2_SB(inode);
495 int error = -EOPNOTSUPP;
496 struct buffer_head *dibh;
497 struct gfs2_alloc *al = &ip->i_alloc;
498 struct gfs2_dinode *di;
500 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
503 error = gfs2_meta_inode_buffer(ip, &dibh);
507 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
508 di = (struct gfs2_dinode *)dibh->b_data;
510 if (gfs2_is_stuffed(ip)) {
514 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
516 kaddr = kmap_atomic(page, KM_USER0);
517 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
518 kaddr + from, to - from);
519 kunmap_atomic(kaddr, KM_USER0);
521 SetPageUptodate(page);
523 if (inode->i_size < file_size) {
524 i_size_write(inode, file_size);
525 mark_inode_dirty(inode);
528 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
530 gfs2_page_add_databufs(ip, page, from, to);
531 error = generic_commit_write(file, page, from, to);
536 if (ip->i_di.di_size < inode->i_size) {
537 ip->i_di.di_size = inode->i_size;
538 di->di_size = cpu_to_be64(inode->i_size);
541 if (inode == sdp->sd_rindex)
542 adjust_fs_space(inode);
546 if (al->al_requested) {
547 gfs2_inplace_release(ip);
548 gfs2_quota_unlock(ip);
552 gfs2_glock_dq_m(1, &ip->i_gh);
554 gfs2_holder_uninit(&ip->i_gh);
561 if (al->al_requested) {
562 gfs2_inplace_release(ip);
563 gfs2_quota_unlock(ip);
567 gfs2_glock_dq_m(1, &ip->i_gh);
569 gfs2_holder_uninit(&ip->i_gh);
571 ClearPageUptodate(page);
576 * gfs2_set_page_dirty - Page dirtying function
577 * @page: The page to dirty
579 * Returns: 1 if it dirtyed the page, or 0 otherwise
582 static int gfs2_set_page_dirty(struct page *page)
584 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
585 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
587 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
588 SetPageChecked(page);
589 return __set_page_dirty_buffers(page);
593 * gfs2_bmap - Block map function
594 * @mapping: Address space info
595 * @lblock: The block to map
597 * Returns: The disk address for the block or 0 on hole or error
600 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
602 struct gfs2_inode *ip = GFS2_I(mapping->host);
603 struct gfs2_holder i_gh;
607 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
611 if (!gfs2_is_stuffed(ip))
612 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
614 gfs2_glock_dq_uninit(&i_gh);
619 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
621 BUG_ON(!PageLocked(page));
623 ClearPageChecked(page);
625 block_invalidatepage(page, offset);
629 * gfs2_ok_for_dio - check that dio is valid on this file
632 * @offset: The offset at which we are reading or writing
634 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
635 * 1 (to accept the i/o request)
637 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
640 * Should we return an error here? I can't see that O_DIRECT for
641 * a journaled file makes any sense. For now we'll silently fall
642 * back to buffered I/O, likewise we do the same for stuffed
643 * files since they are (a) small and (b) unaligned.
645 if (gfs2_is_jdata(ip))
648 if (gfs2_is_stuffed(ip))
651 if (offset > i_size_read(&ip->i_inode))
658 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
659 const struct iovec *iov, loff_t offset,
660 unsigned long nr_segs)
662 struct file *file = iocb->ki_filp;
663 struct inode *inode = file->f_mapping->host;
664 struct gfs2_inode *ip = GFS2_I(inode);
665 struct gfs2_holder gh;
669 * Deferred lock, even if its a write, since we do no allocation
670 * on this path. All we need change is atime, and this lock mode
671 * ensures that other nodes have flushed their buffered read caches
672 * (i.e. their page cache entries for this inode). We do not,
673 * unfortunately have the option of only flushing a range like
676 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
677 rv = gfs2_glock_nq_atime(&gh);
680 rv = gfs2_ok_for_dio(ip, rw, offset);
682 goto out; /* dio not valid, fall back to buffered i/o */
684 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
685 iov, offset, nr_segs,
686 gfs2_get_block_direct, NULL);
688 gfs2_glock_dq_m(1, &gh);
689 gfs2_holder_uninit(&gh);
694 * gfs2_releasepage - free the metadata associated with a page
695 * @page: the page that's being released
696 * @gfp_mask: passed from Linux VFS, ignored by us
698 * Call try_to_free_buffers() if the buffers in this page can be
704 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
706 struct inode *aspace = page->mapping->host;
707 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
708 struct buffer_head *bh, *head;
709 struct gfs2_bufdata *bd;
711 if (!page_has_buffers(page))
715 head = bh = page_buffers(page);
717 if (atomic_read(&bh->b_count))
720 if (bd && bd->bd_ail)
722 gfs2_assert_warn(sdp, !buffer_pinned(bh));
723 gfs2_assert_warn(sdp, !buffer_dirty(bh));
724 bh = bh->b_this_page;
726 gfs2_log_unlock(sdp);
728 head = bh = page_buffers(page);
733 gfs2_assert_warn(sdp, bd->bd_bh == bh);
734 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
736 if (!list_empty(&bd->bd_le.le_list))
738 bh->b_private = NULL;
740 gfs2_log_unlock(sdp);
742 kmem_cache_free(gfs2_bufdata_cachep, bd);
744 bh = bh->b_this_page;
745 } while (bh != head);
748 return try_to_free_buffers(page);
750 gfs2_log_unlock(sdp);
754 const struct address_space_operations gfs2_file_aops = {
755 .writepage = gfs2_writepage,
756 .writepages = gfs2_writepages,
757 .readpage = gfs2_readpage,
758 .readpages = gfs2_readpages,
759 .sync_page = block_sync_page,
760 .prepare_write = gfs2_prepare_write,
761 .commit_write = gfs2_commit_write,
762 .set_page_dirty = gfs2_set_page_dirty,
764 .invalidatepage = gfs2_invalidatepage,
765 .releasepage = gfs2_releasepage,
766 .direct_IO = gfs2_direct_IO,