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/swap.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/lm_interface.h>
31 #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 - 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 int gfs2_get_block(struct inode *inode, sector_t lblock,
71 struct buffer_head *bh_result, int create)
73 return gfs2_block_map(inode, lblock, create, bh_result);
77 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
79 * @lblock: The block number to look up
80 * @bh_result: The buffer head to return the result in
81 * @create: Non-zero if we may add block to the file
86 static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
87 struct buffer_head *bh_result, int create)
91 error = gfs2_block_map(inode, lblock, 0, bh_result);
94 if (!buffer_mapped(bh_result))
99 static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
100 struct buffer_head *bh_result, int create)
102 return gfs2_block_map(inode, lblock, 0, bh_result);
106 * gfs2_writepage - Write complete page
107 * @page: Page to write
111 * Some of this is copied from block_write_full_page() although we still
112 * call it to do most of the work.
115 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
117 struct inode *inode = page->mapping->host;
118 struct gfs2_inode *ip = GFS2_I(inode);
119 struct gfs2_sbd *sdp = GFS2_SB(inode);
120 loff_t i_size = i_size_read(inode);
121 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
126 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
130 if (current->journal_info)
133 /* Is the page fully outside i_size? (truncate in progress) */
134 offset = i_size & (PAGE_CACHE_SIZE-1);
135 if (page->index > end_index || (page->index == end_index && !offset)) {
136 page->mapping->a_ops->invalidatepage(page, 0);
138 return 0; /* don't care */
141 if ((sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) &&
143 ClearPageChecked(page);
144 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
147 if (!page_has_buffers(page)) {
148 create_empty_buffers(page, inode->i_sb->s_blocksize,
149 (1 << BH_Dirty)|(1 << BH_Uptodate));
151 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
154 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
157 gfs2_meta_cache_flush(ip);
161 redirty_page_for_writepage(wbc, page);
167 * gfs2_writepages - Write a bunch of dirty pages back to disk
168 * @mapping: The mapping to write
169 * @wbc: Write-back control
171 * For journaled files and/or ordered writes this just falls back to the
172 * kernel's default writepages path for now. We will probably want to change
173 * that eventually (i.e. when we look at allocate on flush).
175 * For the data=writeback case though we can already ignore buffer heads
176 * and write whole extents at once. This is a big reduction in the
177 * number of I/O requests we send and the bmap calls we make in this case.
179 static int gfs2_writepages(struct address_space *mapping,
180 struct writeback_control *wbc)
182 struct inode *inode = mapping->host;
183 struct gfs2_inode *ip = GFS2_I(inode);
184 struct gfs2_sbd *sdp = GFS2_SB(inode);
186 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
187 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
189 return generic_writepages(mapping, wbc);
193 * stuffed_readpage - Fill in a Linux page with stuffed file data
200 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
202 struct buffer_head *dibh;
207 * Due to the order of unstuffing files and ->nopage(), we can be
208 * asked for a zero page in the case of a stuffed file being extended,
209 * so we need to supply one here. It doesn't happen often.
211 if (unlikely(page->index)) {
212 zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
216 error = gfs2_meta_inode_buffer(ip, &dibh);
220 kaddr = kmap_atomic(page, KM_USER0);
221 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
223 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
224 kunmap_atomic(kaddr, KM_USER0);
225 flush_dcache_page(page);
227 SetPageUptodate(page);
234 * gfs2_readpage - readpage with locking
235 * @file: The file to read a page for. N.B. This may be NULL if we are
236 * reading an internal file.
237 * @page: The page to read
242 static int gfs2_readpage(struct file *file, struct page *page)
244 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
245 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
246 struct gfs2_file *gf = NULL;
247 struct gfs2_holder gh;
251 if (likely(file != &gfs2_internal_file_sentinel)) {
253 gf = file->private_data;
254 if (test_bit(GFF_EXLOCK, &gf->f_flags))
255 /* gfs2_sharewrite_fault has grabbed the ip->i_gl already */
258 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
260 error = gfs2_glock_nq_atime(&gh);
266 if (gfs2_is_stuffed(ip)) {
267 error = stuffed_readpage(ip, page);
270 error = mpage_readpage(page, gfs2_get_block);
272 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
276 gfs2_glock_dq_m(1, &gh);
277 gfs2_holder_uninit(&gh);
283 if (error == GLR_TRYFAILED) {
284 error = AOP_TRUNCATED_PAGE;
288 gfs2_holder_uninit(&gh);
293 * gfs2_readpages - Read a bunch of pages at once
296 * 1. This is only for readahead, so we can simply ignore any things
297 * which are slightly inconvenient (such as locking conflicts between
298 * the page lock and the glock) and return having done no I/O. Its
299 * obviously not something we'd want to do on too regular a basis.
300 * Any I/O we ignore at this time will be done via readpage later.
301 * 2. We don't handle stuffed files here we let readpage do the honours.
302 * 3. mpage_readpages() does most of the heavy lifting in the common case.
303 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
304 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
305 * well as read-ahead.
307 static int gfs2_readpages(struct file *file, struct address_space *mapping,
308 struct list_head *pages, unsigned nr_pages)
310 struct inode *inode = mapping->host;
311 struct gfs2_inode *ip = GFS2_I(inode);
312 struct gfs2_sbd *sdp = GFS2_SB(inode);
313 struct gfs2_holder gh;
317 if (likely(file != &gfs2_internal_file_sentinel)) {
319 struct gfs2_file *gf = file->private_data;
320 if (test_bit(GFF_EXLOCK, &gf->f_flags))
323 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
324 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
326 ret = gfs2_glock_nq_atime(&gh);
327 if (ret == GLR_TRYFAILED)
333 if (!gfs2_is_stuffed(ip))
334 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
337 gfs2_glock_dq_m(1, &gh);
338 gfs2_holder_uninit(&gh);
341 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
348 gfs2_holder_uninit(&gh);
353 * gfs2_write_begin - Begin to write to a file
354 * @file: The file to write to
355 * @mapping: The mapping in which to write
356 * @pos: The file offset at which to start writing
357 * @len: Length of the write
358 * @flags: Various flags
359 * @pagep: Pointer to return the page
360 * @fsdata: Pointer to return fs data (unused by GFS2)
365 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
366 loff_t pos, unsigned len, unsigned flags,
367 struct page **pagep, void **fsdata)
369 struct gfs2_inode *ip = GFS2_I(mapping->host);
370 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
371 unsigned int data_blocks, ind_blocks, rblocks;
374 struct gfs2_alloc *al;
375 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
376 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
377 unsigned to = from + len;
380 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
381 error = gfs2_glock_nq_atime(&ip->i_gh);
386 page = __grab_cache_page(mapping, index);
391 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
393 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
398 ip->i_alloc.al_requested = 0;
399 if (alloc_required) {
400 al = gfs2_alloc_get(ip);
402 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
406 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
410 al->al_requested = data_blocks + ind_blocks;
411 error = gfs2_inplace_reserve(ip);
416 rblocks = RES_DINODE + ind_blocks;
417 if (gfs2_is_jdata(ip))
418 rblocks += data_blocks ? data_blocks : 1;
419 if (ind_blocks || data_blocks)
420 rblocks += RES_STATFS + RES_QUOTA;
422 error = gfs2_trans_begin(sdp, rblocks,
423 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
427 if (gfs2_is_stuffed(ip)) {
428 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
429 error = gfs2_unstuff_dinode(ip, page);
432 } else if (!PageUptodate(page))
433 error = stuffed_readpage(ip, page);
438 error = block_prepare_write(page, from, to, gfs2_get_block);
444 if (alloc_required) {
445 gfs2_inplace_release(ip);
447 gfs2_quota_unlock(ip);
452 page_cache_release(page);
453 if (pos + len > ip->i_inode.i_size)
454 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
456 gfs2_glock_dq_m(1, &ip->i_gh);
458 gfs2_holder_uninit(&ip->i_gh);
465 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
466 * @inode: the rindex inode
468 static void adjust_fs_space(struct inode *inode)
470 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
471 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
472 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
473 u64 fs_total, new_free;
475 /* Total up the file system space, according to the latest rindex. */
476 fs_total = gfs2_ri_total(sdp);
478 spin_lock(&sdp->sd_statfs_spin);
479 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
480 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
483 spin_unlock(&sdp->sd_statfs_spin);
484 fs_warn(sdp, "File system extended by %llu blocks.\n",
485 (unsigned long long)new_free);
486 gfs2_statfs_change(sdp, new_free, new_free, 0);
490 * gfs2_stuffed_write_end - Write end for stuffed files
492 * @dibh: The buffer_head containing the on-disk inode
493 * @pos: The file position
494 * @len: The length of the write
495 * @copied: How much was actually copied by the VFS
498 * This copies the data from the page into the inode block after
499 * the inode data structure itself.
503 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
504 loff_t pos, unsigned len, unsigned copied,
507 struct gfs2_inode *ip = GFS2_I(inode);
508 struct gfs2_sbd *sdp = GFS2_SB(inode);
509 u64 to = pos + copied;
511 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
512 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
514 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
515 kaddr = kmap_atomic(page, KM_USER0);
516 memcpy(buf + pos, kaddr + pos, copied);
517 memset(kaddr + pos + copied, 0, len - copied);
518 flush_dcache_page(page);
519 kunmap_atomic(kaddr, KM_USER0);
521 if (!PageUptodate(page))
522 SetPageUptodate(page);
524 page_cache_release(page);
526 if (inode->i_size < to) {
527 i_size_write(inode, to);
528 ip->i_di.di_size = inode->i_size;
529 di->di_size = cpu_to_be64(inode->i_size);
530 mark_inode_dirty(inode);
533 if (inode == sdp->sd_rindex)
534 adjust_fs_space(inode);
538 gfs2_glock_dq(&ip->i_gh);
539 gfs2_holder_uninit(&ip->i_gh);
545 * @file: The file to write to
546 * @mapping: The address space to write to
547 * @pos: The file position
548 * @len: The length of the data
550 * @page: The page that has been written
551 * @fsdata: The fsdata (unused in GFS2)
553 * The main write_end function for GFS2. We have a separate one for
554 * stuffed files as they are slightly different, otherwise we just
555 * put our locking around the VFS provided functions.
560 static int gfs2_write_end(struct file *file, struct address_space *mapping,
561 loff_t pos, unsigned len, unsigned copied,
562 struct page *page, void *fsdata)
564 struct inode *inode = page->mapping->host;
565 struct gfs2_inode *ip = GFS2_I(inode);
566 struct gfs2_sbd *sdp = GFS2_SB(inode);
567 struct buffer_head *dibh;
568 struct gfs2_alloc *al = &ip->i_alloc;
569 struct gfs2_dinode *di;
570 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
571 unsigned int to = from + len;
574 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == 0);
576 ret = gfs2_meta_inode_buffer(ip, &dibh);
579 page_cache_release(page);
583 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
585 if (gfs2_is_stuffed(ip))
586 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
588 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
589 gfs2_page_add_databufs(ip, page, from, to);
591 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
593 if (likely(ret >= 0)) {
595 if ((pos + copied) > inode->i_size) {
596 di = (struct gfs2_dinode *)dibh->b_data;
597 ip->i_di.di_size = inode->i_size;
598 di->di_size = cpu_to_be64(inode->i_size);
599 mark_inode_dirty(inode);
603 if (inode == sdp->sd_rindex)
604 adjust_fs_space(inode);
609 if (al->al_requested) {
610 gfs2_inplace_release(ip);
611 gfs2_quota_unlock(ip);
614 gfs2_glock_dq(&ip->i_gh);
615 gfs2_holder_uninit(&ip->i_gh);
620 * gfs2_set_page_dirty - Page dirtying function
621 * @page: The page to dirty
623 * Returns: 1 if it dirtyed the page, or 0 otherwise
626 static int gfs2_set_page_dirty(struct page *page)
628 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
629 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
631 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
632 SetPageChecked(page);
633 return __set_page_dirty_buffers(page);
637 * gfs2_bmap - Block map function
638 * @mapping: Address space info
639 * @lblock: The block to map
641 * Returns: The disk address for the block or 0 on hole or error
644 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
646 struct gfs2_inode *ip = GFS2_I(mapping->host);
647 struct gfs2_holder i_gh;
651 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
655 if (!gfs2_is_stuffed(ip))
656 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
658 gfs2_glock_dq_uninit(&i_gh);
663 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
665 struct gfs2_bufdata *bd;
669 clear_buffer_dirty(bh);
672 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
673 list_del_init(&bd->bd_le.le_list);
675 gfs2_remove_from_journal(bh, current->journal_info, 0);
678 clear_buffer_mapped(bh);
679 clear_buffer_req(bh);
680 clear_buffer_new(bh);
681 gfs2_log_unlock(sdp);
685 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
687 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
688 struct buffer_head *bh, *head;
689 unsigned long pos = 0;
691 BUG_ON(!PageLocked(page));
693 ClearPageChecked(page);
694 if (!page_has_buffers(page))
697 bh = head = page_buffers(page);
700 gfs2_discard(sdp, bh);
702 bh = bh->b_this_page;
703 } while (bh != head);
706 try_to_release_page(page, 0);
710 * gfs2_ok_for_dio - check that dio is valid on this file
713 * @offset: The offset at which we are reading or writing
715 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
716 * 1 (to accept the i/o request)
718 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
721 * Should we return an error here? I can't see that O_DIRECT for
722 * a journaled file makes any sense. For now we'll silently fall
723 * back to buffered I/O, likewise we do the same for stuffed
724 * files since they are (a) small and (b) unaligned.
726 if (gfs2_is_jdata(ip))
729 if (gfs2_is_stuffed(ip))
732 if (offset > i_size_read(&ip->i_inode))
739 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
740 const struct iovec *iov, loff_t offset,
741 unsigned long nr_segs)
743 struct file *file = iocb->ki_filp;
744 struct inode *inode = file->f_mapping->host;
745 struct gfs2_inode *ip = GFS2_I(inode);
746 struct gfs2_holder gh;
750 * Deferred lock, even if its a write, since we do no allocation
751 * on this path. All we need change is atime, and this lock mode
752 * ensures that other nodes have flushed their buffered read caches
753 * (i.e. their page cache entries for this inode). We do not,
754 * unfortunately have the option of only flushing a range like
757 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
758 rv = gfs2_glock_nq_atime(&gh);
761 rv = gfs2_ok_for_dio(ip, rw, offset);
763 goto out; /* dio not valid, fall back to buffered i/o */
765 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
766 iov, offset, nr_segs,
767 gfs2_get_block_direct, NULL);
769 gfs2_glock_dq_m(1, &gh);
770 gfs2_holder_uninit(&gh);
775 * gfs2_releasepage - free the metadata associated with a page
776 * @page: the page that's being released
777 * @gfp_mask: passed from Linux VFS, ignored by us
779 * Call try_to_free_buffers() if the buffers in this page can be
785 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
787 struct inode *aspace = page->mapping->host;
788 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
789 struct buffer_head *bh, *head;
790 struct gfs2_bufdata *bd;
792 if (!page_has_buffers(page))
796 head = bh = page_buffers(page);
798 if (atomic_read(&bh->b_count))
801 if (bd && bd->bd_ail)
803 gfs2_assert_warn(sdp, !buffer_pinned(bh));
804 gfs2_assert_warn(sdp, !buffer_dirty(bh));
805 bh = bh->b_this_page;
807 gfs2_log_unlock(sdp);
809 head = bh = page_buffers(page);
814 gfs2_assert_warn(sdp, bd->bd_bh == bh);
815 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
816 if (!list_empty(&bd->bd_le.le_list)) {
817 if (!buffer_pinned(bh))
818 list_del_init(&bd->bd_le.le_list);
824 bh->b_private = NULL;
826 gfs2_log_unlock(sdp);
828 kmem_cache_free(gfs2_bufdata_cachep, bd);
830 bh = bh->b_this_page;
831 } while (bh != head);
833 return try_to_free_buffers(page);
835 gfs2_log_unlock(sdp);
839 const struct address_space_operations gfs2_file_aops = {
840 .writepage = gfs2_writepage,
841 .writepages = gfs2_writepages,
842 .readpage = gfs2_readpage,
843 .readpages = gfs2_readpages,
844 .sync_page = block_sync_page,
845 .write_begin = gfs2_write_begin,
846 .write_end = gfs2_write_end,
847 .set_page_dirty = gfs2_set_page_dirty,
849 .invalidatepage = gfs2_invalidatepage,
850 .releasepage = gfs2_releasepage,
851 .direct_IO = gfs2_direct_IO,