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>
23 #include <linux/swap.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 - 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 (PageChecked(page)) {
142 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
145 ClearPageChecked(page);
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);
159 redirty_page_for_writepage(wbc, page);
165 * gfs2_writepages - Write a bunch of dirty pages back to disk
166 * @mapping: The mapping to write
167 * @wbc: Write-back control
169 * For journaled files and/or ordered writes this just falls back to the
170 * kernel's default writepages path for now. We will probably want to change
171 * that eventually (i.e. when we look at allocate on flush).
173 * For the data=writeback case though we can already ignore buffer heads
174 * and write whole extents at once. This is a big reduction in the
175 * number of I/O requests we send and the bmap calls we make in this case.
177 static int gfs2_writepages(struct address_space *mapping,
178 struct writeback_control *wbc)
180 struct inode *inode = mapping->host;
181 struct gfs2_inode *ip = GFS2_I(inode);
183 if (gfs2_is_writeback(ip))
184 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
186 return generic_writepages(mapping, wbc);
190 * stuffed_readpage - Fill in a Linux page with stuffed file data
197 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
199 struct buffer_head *dibh;
204 * Due to the order of unstuffing files and ->nopage(), we can be
205 * asked for a zero page in the case of a stuffed file being extended,
206 * so we need to supply one here. It doesn't happen often.
208 if (unlikely(page->index)) {
209 zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
213 error = gfs2_meta_inode_buffer(ip, &dibh);
217 kaddr = kmap_atomic(page, KM_USER0);
218 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
220 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
221 kunmap_atomic(kaddr, KM_USER0);
222 flush_dcache_page(page);
224 SetPageUptodate(page);
231 * __gfs2_readpage - readpage
232 * @file: The file to read a page for
233 * @page: The page to read
235 * This is the core of gfs2's readpage. Its used by the internal file
236 * reading code as in that case we already hold the glock. Also its
237 * called by gfs2_readpage() once the required lock has been granted.
241 static int __gfs2_readpage(void *file, struct page *page)
243 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
244 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
247 if (gfs2_is_stuffed(ip)) {
248 error = stuffed_readpage(ip, page);
251 error = mpage_readpage(page, gfs2_get_block);
254 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
261 * gfs2_readpage - read a page of a file
262 * @file: The file to read
263 * @page: The page of the file
265 * This deals with the locking required. We use a trylock in order to
266 * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
267 * in the event that we are unable to get the lock.
270 static int gfs2_readpage(struct file *file, struct page *page)
272 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
273 struct gfs2_holder gh;
276 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
277 error = gfs2_glock_nq_atime(&gh);
278 if (unlikely(error)) {
282 error = __gfs2_readpage(file, page);
285 gfs2_holder_uninit(&gh);
286 if (error == GLR_TRYFAILED) {
288 return AOP_TRUNCATED_PAGE;
294 * gfs2_internal_read - read an internal file
295 * @ip: The gfs2 inode
296 * @ra_state: The readahead state (or NULL for no readahead)
297 * @buf: The buffer to fill
298 * @pos: The file position
299 * @size: The amount to read
303 int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state,
304 char *buf, loff_t *pos, unsigned size)
306 struct address_space *mapping = ip->i_inode.i_mapping;
307 unsigned long index = *pos / PAGE_CACHE_SIZE;
308 unsigned offset = *pos & (PAGE_CACHE_SIZE - 1);
316 if (offset + size > PAGE_CACHE_SIZE)
317 amt = PAGE_CACHE_SIZE - offset;
318 page = read_cache_page(mapping, index, __gfs2_readpage, NULL);
320 return PTR_ERR(page);
321 p = kmap_atomic(page, KM_USER0);
322 memcpy(buf + copied, p + offset, amt);
323 kunmap_atomic(p, KM_USER0);
324 mark_page_accessed(page);
325 page_cache_release(page);
329 } while(copied < size);
335 * gfs2_readpages - Read a bunch of pages at once
338 * 1. This is only for readahead, so we can simply ignore any things
339 * which are slightly inconvenient (such as locking conflicts between
340 * the page lock and the glock) and return having done no I/O. Its
341 * obviously not something we'd want to do on too regular a basis.
342 * Any I/O we ignore at this time will be done via readpage later.
343 * 2. We don't handle stuffed files here we let readpage do the honours.
344 * 3. mpage_readpages() does most of the heavy lifting in the common case.
345 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
348 static int gfs2_readpages(struct file *file, struct address_space *mapping,
349 struct list_head *pages, unsigned nr_pages)
351 struct inode *inode = mapping->host;
352 struct gfs2_inode *ip = GFS2_I(inode);
353 struct gfs2_sbd *sdp = GFS2_SB(inode);
354 struct gfs2_holder gh;
357 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
358 ret = gfs2_glock_nq_atime(&gh);
361 if (!gfs2_is_stuffed(ip))
362 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
365 gfs2_holder_uninit(&gh);
366 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
372 * gfs2_write_begin - Begin to write to a file
373 * @file: The file to write to
374 * @mapping: The mapping in which to write
375 * @pos: The file offset at which to start writing
376 * @len: Length of the write
377 * @flags: Various flags
378 * @pagep: Pointer to return the page
379 * @fsdata: Pointer to return fs data (unused by GFS2)
384 static int gfs2_write_begin(struct file *file, struct address_space *mapping,
385 loff_t pos, unsigned len, unsigned flags,
386 struct page **pagep, void **fsdata)
388 struct gfs2_inode *ip = GFS2_I(mapping->host);
389 struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
390 unsigned int data_blocks, ind_blocks, rblocks;
393 struct gfs2_alloc *al;
394 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
395 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
396 unsigned to = from + len;
399 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
400 error = gfs2_glock_nq_atime(&ip->i_gh);
405 page = __grab_cache_page(mapping, index);
410 gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
412 error = gfs2_write_alloc_required(ip, pos, len, &alloc_required);
417 ip->i_alloc.al_requested = 0;
418 if (alloc_required) {
419 al = gfs2_alloc_get(ip);
421 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
425 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
429 al->al_requested = data_blocks + ind_blocks;
430 error = gfs2_inplace_reserve(ip);
435 rblocks = RES_DINODE + ind_blocks;
436 if (gfs2_is_jdata(ip))
437 rblocks += data_blocks ? data_blocks : 1;
438 if (ind_blocks || data_blocks)
439 rblocks += RES_STATFS + RES_QUOTA;
441 error = gfs2_trans_begin(sdp, rblocks,
442 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
446 if (gfs2_is_stuffed(ip)) {
447 if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
448 error = gfs2_unstuff_dinode(ip, page);
451 } else if (!PageUptodate(page))
452 error = stuffed_readpage(ip, page);
457 error = block_prepare_write(page, from, to, gfs2_get_block);
463 if (alloc_required) {
464 gfs2_inplace_release(ip);
466 gfs2_quota_unlock(ip);
471 page_cache_release(page);
472 if (pos + len > ip->i_inode.i_size)
473 vmtruncate(&ip->i_inode, ip->i_inode.i_size);
475 gfs2_glock_dq_m(1, &ip->i_gh);
477 gfs2_holder_uninit(&ip->i_gh);
484 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
485 * @inode: the rindex inode
487 static void adjust_fs_space(struct inode *inode)
489 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
490 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
491 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
492 u64 fs_total, new_free;
494 /* Total up the file system space, according to the latest rindex. */
495 fs_total = gfs2_ri_total(sdp);
497 spin_lock(&sdp->sd_statfs_spin);
498 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
499 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
502 spin_unlock(&sdp->sd_statfs_spin);
503 fs_warn(sdp, "File system extended by %llu blocks.\n",
504 (unsigned long long)new_free);
505 gfs2_statfs_change(sdp, new_free, new_free, 0);
509 * gfs2_stuffed_write_end - Write end for stuffed files
511 * @dibh: The buffer_head containing the on-disk inode
512 * @pos: The file position
513 * @len: The length of the write
514 * @copied: How much was actually copied by the VFS
517 * This copies the data from the page into the inode block after
518 * the inode data structure itself.
522 static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
523 loff_t pos, unsigned len, unsigned copied,
526 struct gfs2_inode *ip = GFS2_I(inode);
527 struct gfs2_sbd *sdp = GFS2_SB(inode);
528 u64 to = pos + copied;
530 unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode);
531 struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data;
533 BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode)));
534 kaddr = kmap_atomic(page, KM_USER0);
535 memcpy(buf + pos, kaddr + pos, copied);
536 memset(kaddr + pos + copied, 0, len - copied);
537 flush_dcache_page(page);
538 kunmap_atomic(kaddr, KM_USER0);
540 if (!PageUptodate(page))
541 SetPageUptodate(page);
543 page_cache_release(page);
545 if (inode->i_size < to) {
546 i_size_write(inode, to);
547 ip->i_di.di_size = inode->i_size;
548 di->di_size = cpu_to_be64(inode->i_size);
549 mark_inode_dirty(inode);
552 if (inode == sdp->sd_rindex)
553 adjust_fs_space(inode);
557 gfs2_glock_dq(&ip->i_gh);
558 gfs2_holder_uninit(&ip->i_gh);
564 * @file: The file to write to
565 * @mapping: The address space to write to
566 * @pos: The file position
567 * @len: The length of the data
569 * @page: The page that has been written
570 * @fsdata: The fsdata (unused in GFS2)
572 * The main write_end function for GFS2. We have a separate one for
573 * stuffed files as they are slightly different, otherwise we just
574 * put our locking around the VFS provided functions.
579 static int gfs2_write_end(struct file *file, struct address_space *mapping,
580 loff_t pos, unsigned len, unsigned copied,
581 struct page *page, void *fsdata)
583 struct inode *inode = page->mapping->host;
584 struct gfs2_inode *ip = GFS2_I(inode);
585 struct gfs2_sbd *sdp = GFS2_SB(inode);
586 struct buffer_head *dibh;
587 struct gfs2_alloc *al = &ip->i_alloc;
588 struct gfs2_dinode *di;
589 unsigned int from = pos & (PAGE_CACHE_SIZE - 1);
590 unsigned int to = from + len;
593 BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == 0);
595 ret = gfs2_meta_inode_buffer(ip, &dibh);
598 page_cache_release(page);
602 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
604 if (gfs2_is_stuffed(ip))
605 return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page);
607 if (!gfs2_is_writeback(ip))
608 gfs2_page_add_databufs(ip, page, from, to);
610 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
612 if (likely(ret >= 0)) {
614 if ((pos + copied) > inode->i_size) {
615 di = (struct gfs2_dinode *)dibh->b_data;
616 ip->i_di.di_size = inode->i_size;
617 di->di_size = cpu_to_be64(inode->i_size);
618 mark_inode_dirty(inode);
622 if (inode == sdp->sd_rindex)
623 adjust_fs_space(inode);
628 if (al->al_requested) {
629 gfs2_inplace_release(ip);
630 gfs2_quota_unlock(ip);
633 gfs2_glock_dq(&ip->i_gh);
634 gfs2_holder_uninit(&ip->i_gh);
639 * gfs2_set_page_dirty - Page dirtying function
640 * @page: The page to dirty
642 * Returns: 1 if it dirtyed the page, or 0 otherwise
645 static int gfs2_set_page_dirty(struct page *page)
647 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
648 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
650 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
651 SetPageChecked(page);
652 return __set_page_dirty_buffers(page);
656 * gfs2_bmap - Block map function
657 * @mapping: Address space info
658 * @lblock: The block to map
660 * Returns: The disk address for the block or 0 on hole or error
663 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
665 struct gfs2_inode *ip = GFS2_I(mapping->host);
666 struct gfs2_holder i_gh;
670 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
674 if (!gfs2_is_stuffed(ip))
675 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
677 gfs2_glock_dq_uninit(&i_gh);
682 static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh)
684 struct gfs2_bufdata *bd;
688 clear_buffer_dirty(bh);
691 if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh))
692 list_del_init(&bd->bd_le.le_list);
694 gfs2_remove_from_journal(bh, current->journal_info, 0);
697 clear_buffer_mapped(bh);
698 clear_buffer_req(bh);
699 clear_buffer_new(bh);
700 gfs2_log_unlock(sdp);
704 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
706 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
707 struct buffer_head *bh, *head;
708 unsigned long pos = 0;
710 BUG_ON(!PageLocked(page));
712 ClearPageChecked(page);
713 if (!page_has_buffers(page))
716 bh = head = page_buffers(page);
719 gfs2_discard(sdp, bh);
721 bh = bh->b_this_page;
722 } while (bh != head);
725 try_to_release_page(page, 0);
729 * gfs2_ok_for_dio - check that dio is valid on this file
732 * @offset: The offset at which we are reading or writing
734 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
735 * 1 (to accept the i/o request)
737 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
740 * Should we return an error here? I can't see that O_DIRECT for
741 * a journaled file makes any sense. For now we'll silently fall
742 * back to buffered I/O, likewise we do the same for stuffed
743 * files since they are (a) small and (b) unaligned.
745 if (gfs2_is_jdata(ip))
748 if (gfs2_is_stuffed(ip))
751 if (offset > i_size_read(&ip->i_inode))
758 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
759 const struct iovec *iov, loff_t offset,
760 unsigned long nr_segs)
762 struct file *file = iocb->ki_filp;
763 struct inode *inode = file->f_mapping->host;
764 struct gfs2_inode *ip = GFS2_I(inode);
765 struct gfs2_holder gh;
769 * Deferred lock, even if its a write, since we do no allocation
770 * on this path. All we need change is atime, and this lock mode
771 * ensures that other nodes have flushed their buffered read caches
772 * (i.e. their page cache entries for this inode). We do not,
773 * unfortunately have the option of only flushing a range like
776 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
777 rv = gfs2_glock_nq_atime(&gh);
780 rv = gfs2_ok_for_dio(ip, rw, offset);
782 goto out; /* dio not valid, fall back to buffered i/o */
784 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
785 iov, offset, nr_segs,
786 gfs2_get_block_direct, NULL);
788 gfs2_glock_dq_m(1, &gh);
789 gfs2_holder_uninit(&gh);
794 * gfs2_releasepage - free the metadata associated with a page
795 * @page: the page that's being released
796 * @gfp_mask: passed from Linux VFS, ignored by us
798 * Call try_to_free_buffers() if the buffers in this page can be
804 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
806 struct inode *aspace = page->mapping->host;
807 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
808 struct buffer_head *bh, *head;
809 struct gfs2_bufdata *bd;
811 if (!page_has_buffers(page))
815 head = bh = page_buffers(page);
817 if (atomic_read(&bh->b_count))
820 if (bd && bd->bd_ail)
822 gfs2_assert_warn(sdp, !buffer_pinned(bh));
823 gfs2_assert_warn(sdp, !buffer_dirty(bh));
824 bh = bh->b_this_page;
826 gfs2_log_unlock(sdp);
828 head = bh = page_buffers(page);
833 gfs2_assert_warn(sdp, bd->bd_bh == bh);
834 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
835 if (!list_empty(&bd->bd_le.le_list)) {
836 if (!buffer_pinned(bh))
837 list_del_init(&bd->bd_le.le_list);
843 bh->b_private = NULL;
845 gfs2_log_unlock(sdp);
847 kmem_cache_free(gfs2_bufdata_cachep, bd);
849 bh = bh->b_this_page;
850 } while (bh != head);
852 return try_to_free_buffers(page);
854 gfs2_log_unlock(sdp);
858 const struct address_space_operations gfs2_file_aops = {
859 .writepage = gfs2_writepage,
860 .writepages = gfs2_writepages,
861 .readpage = gfs2_readpage,
862 .readpages = gfs2_readpages,
863 .sync_page = block_sync_page,
864 .write_begin = gfs2_write_begin,
865 .write_end = gfs2_write_end,
866 .set_page_dirty = gfs2_set_page_dirty,
868 .invalidatepage = gfs2_invalidatepage,
869 .releasepage = gfs2_releasepage,
870 .direct_IO = gfs2_direct_IO,