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 kaddr = kmap_atomic(page, KM_USER0);
212 memset(kaddr, 0, PAGE_CACHE_SIZE);
213 kunmap_atomic(kaddr, KM_USER0);
214 flush_dcache_page(page);
215 SetPageUptodate(page);
219 error = gfs2_meta_inode_buffer(ip, &dibh);
223 kaddr = kmap_atomic(page, KM_USER0);
224 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
226 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
227 kunmap_atomic(kaddr, KM_USER0);
228 flush_dcache_page(page);
230 SetPageUptodate(page);
237 * gfs2_readpage - readpage with locking
238 * @file: The file to read a page for. N.B. This may be NULL if we are
239 * reading an internal file.
240 * @page: The page to read
245 static int gfs2_readpage(struct file *file, struct page *page)
247 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
248 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
249 struct gfs2_file *gf = NULL;
250 struct gfs2_holder gh;
254 if (likely(file != &gfs2_internal_file_sentinel)) {
256 gf = file->private_data;
257 if (test_bit(GFF_EXLOCK, &gf->f_flags))
258 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
261 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
263 error = gfs2_glock_nq_atime(&gh);
269 if (gfs2_is_stuffed(ip)) {
270 error = stuffed_readpage(ip, page);
273 error = mpage_readpage(page, gfs2_get_block);
275 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
279 gfs2_glock_dq_m(1, &gh);
280 gfs2_holder_uninit(&gh);
286 if (error == GLR_TRYFAILED) {
287 error = AOP_TRUNCATED_PAGE;
291 gfs2_holder_uninit(&gh);
296 * gfs2_readpages - Read a bunch of pages at once
299 * 1. This is only for readahead, so we can simply ignore any things
300 * which are slightly inconvenient (such as locking conflicts between
301 * the page lock and the glock) and return having done no I/O. Its
302 * obviously not something we'd want to do on too regular a basis.
303 * Any I/O we ignore at this time will be done via readpage later.
304 * 2. We don't handle stuffed files here we let readpage do the honours.
305 * 3. mpage_readpages() does most of the heavy lifting in the common case.
306 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
307 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
308 * well as read-ahead.
310 static int gfs2_readpages(struct file *file, struct address_space *mapping,
311 struct list_head *pages, unsigned nr_pages)
313 struct inode *inode = mapping->host;
314 struct gfs2_inode *ip = GFS2_I(inode);
315 struct gfs2_sbd *sdp = GFS2_SB(inode);
316 struct gfs2_holder gh;
320 if (likely(file != &gfs2_internal_file_sentinel)) {
322 struct gfs2_file *gf = file->private_data;
323 if (test_bit(GFF_EXLOCK, &gf->f_flags))
326 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
327 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
329 ret = gfs2_glock_nq_atime(&gh);
330 if (ret == GLR_TRYFAILED)
336 if (!gfs2_is_stuffed(ip))
337 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
340 gfs2_glock_dq_m(1, &gh);
341 gfs2_holder_uninit(&gh);
344 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
351 gfs2_holder_uninit(&gh);
356 * gfs2_prepare_write - Prepare to write a page to a file
357 * @file: The file to write to
358 * @page: The page which is to be prepared for writing
359 * @from: From (byte range within page)
360 * @to: To (byte range within page)
365 static int gfs2_prepare_write(struct file *file, struct page *page,
366 unsigned from, unsigned to)
368 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
369 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
370 unsigned int data_blocks, ind_blocks, rblocks;
373 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
374 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
375 struct gfs2_alloc *al;
376 unsigned int write_len = to - from;
379 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
380 error = gfs2_glock_nq_atime(&ip->i_gh);
381 if (unlikely(error)) {
382 if (error == GLR_TRYFAILED) {
384 error = AOP_TRUNCATED_PAGE;
390 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
392 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
397 ip->i_alloc.al_requested = 0;
398 if (alloc_required) {
399 al = gfs2_alloc_get(ip);
401 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
405 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
409 al->al_requested = data_blocks + ind_blocks;
410 error = gfs2_inplace_reserve(ip);
415 rblocks = RES_DINODE + ind_blocks;
416 if (gfs2_is_jdata(ip))
417 rblocks += data_blocks ? data_blocks : 1;
418 if (ind_blocks || data_blocks)
419 rblocks += RES_STATFS + RES_QUOTA;
421 error = gfs2_trans_begin(sdp, rblocks, 0);
425 if (gfs2_is_stuffed(ip)) {
426 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
427 error = gfs2_unstuff_dinode(ip, page);
430 } else if (!PageUptodate(page))
431 error = stuffed_readpage(ip, page);
436 error = block_prepare_write(page, from, to, gfs2_get_block);
441 if (alloc_required) {
442 gfs2_inplace_release(ip);
444 gfs2_quota_unlock(ip);
449 gfs2_glock_dq_m(1, &ip->i_gh);
451 gfs2_holder_uninit(&ip->i_gh);
458 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
459 * @inode: the rindex inode
461 static void adjust_fs_space(struct inode *inode)
463 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
464 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
465 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
466 u64 fs_total, new_free;
468 /* Total up the file system space, according to the latest rindex. */
469 fs_total = gfs2_ri_total(sdp);
471 spin_lock(&sdp->sd_statfs_spin);
472 if (fs_total > (m_sc->sc_total + l_sc->sc_total))
473 new_free = fs_total - (m_sc->sc_total + l_sc->sc_total);
476 spin_unlock(&sdp->sd_statfs_spin);
477 fs_warn(sdp, "File system extended by %llu blocks.\n",
478 (unsigned long long)new_free);
479 gfs2_statfs_change(sdp, new_free, new_free, 0);
483 * gfs2_commit_write - Commit write to a file
484 * @file: The file to write to
485 * @page: The page containing the data
486 * @from: From (byte range within page)
487 * @to: To (byte range within page)
492 static int gfs2_commit_write(struct file *file, struct page *page,
493 unsigned from, unsigned to)
495 struct inode *inode = page->mapping->host;
496 struct gfs2_inode *ip = GFS2_I(inode);
497 struct gfs2_sbd *sdp = GFS2_SB(inode);
498 int error = -EOPNOTSUPP;
499 struct buffer_head *dibh;
500 struct gfs2_alloc *al = &ip->i_alloc;
501 struct gfs2_dinode *di;
503 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
506 error = gfs2_meta_inode_buffer(ip, &dibh);
510 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
511 di = (struct gfs2_dinode *)dibh->b_data;
513 if (gfs2_is_stuffed(ip)) {
517 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
519 kaddr = kmap_atomic(page, KM_USER0);
520 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
521 kaddr + from, to - from);
522 kunmap_atomic(kaddr, KM_USER0);
524 SetPageUptodate(page);
526 if (inode->i_size < file_size) {
527 i_size_write(inode, file_size);
528 mark_inode_dirty(inode);
531 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
533 gfs2_page_add_databufs(ip, page, from, to);
534 error = generic_commit_write(file, page, from, to);
539 if (ip->i_di.di_size < inode->i_size) {
540 ip->i_di.di_size = inode->i_size;
541 di->di_size = cpu_to_be64(inode->i_size);
544 if (inode == sdp->sd_rindex)
545 adjust_fs_space(inode);
549 if (al->al_requested) {
550 gfs2_inplace_release(ip);
551 gfs2_quota_unlock(ip);
555 gfs2_glock_dq_m(1, &ip->i_gh);
557 gfs2_holder_uninit(&ip->i_gh);
564 if (al->al_requested) {
565 gfs2_inplace_release(ip);
566 gfs2_quota_unlock(ip);
570 gfs2_glock_dq_m(1, &ip->i_gh);
572 gfs2_holder_uninit(&ip->i_gh);
574 ClearPageUptodate(page);
579 * gfs2_set_page_dirty - Page dirtying function
580 * @page: The page to dirty
582 * Returns: 1 if it dirtyed the page, or 0 otherwise
585 static int gfs2_set_page_dirty(struct page *page)
587 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
588 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
590 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
591 SetPageChecked(page);
592 return __set_page_dirty_buffers(page);
596 * gfs2_bmap - Block map function
597 * @mapping: Address space info
598 * @lblock: The block to map
600 * Returns: The disk address for the block or 0 on hole or error
603 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
605 struct gfs2_inode *ip = GFS2_I(mapping->host);
606 struct gfs2_holder i_gh;
610 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
614 if (!gfs2_is_stuffed(ip))
615 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
617 gfs2_glock_dq_uninit(&i_gh);
622 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
624 struct gfs2_bufdata *bd;
630 bh->b_private = NULL;
631 if (!bd->bd_ail && list_empty(&bd->bd_le.le_list))
632 kmem_cache_free(gfs2_bufdata_cachep, bd);
634 gfs2_log_unlock(sdp);
637 clear_buffer_dirty(bh);
639 clear_buffer_mapped(bh);
640 clear_buffer_req(bh);
641 clear_buffer_new(bh);
642 clear_buffer_delay(bh);
646 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
648 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
649 struct buffer_head *head, *bh, *next;
650 unsigned int curr_off = 0;
652 BUG_ON(!PageLocked(page));
654 ClearPageChecked(page);
655 if (!page_has_buffers(page))
658 bh = head = page_buffers(page);
660 unsigned int next_off = curr_off + bh->b_size;
661 next = bh->b_this_page;
663 if (offset <= curr_off)
664 discard_buffer(sdp, bh);
668 } while (bh != head);
671 try_to_release_page(page, 0);
677 * gfs2_ok_for_dio - check that dio is valid on this file
680 * @offset: The offset at which we are reading or writing
682 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
683 * 1 (to accept the i/o request)
685 static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
688 * Should we return an error here? I can't see that O_DIRECT for
689 * a journaled file makes any sense. For now we'll silently fall
690 * back to buffered I/O, likewise we do the same for stuffed
691 * files since they are (a) small and (b) unaligned.
693 if (gfs2_is_jdata(ip))
696 if (gfs2_is_stuffed(ip))
699 if (offset > i_size_read(&ip->i_inode))
706 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
707 const struct iovec *iov, loff_t offset,
708 unsigned long nr_segs)
710 struct file *file = iocb->ki_filp;
711 struct inode *inode = file->f_mapping->host;
712 struct gfs2_inode *ip = GFS2_I(inode);
713 struct gfs2_holder gh;
717 * Deferred lock, even if its a write, since we do no allocation
718 * on this path. All we need change is atime, and this lock mode
719 * ensures that other nodes have flushed their buffered read caches
720 * (i.e. their page cache entries for this inode). We do not,
721 * unfortunately have the option of only flushing a range like
724 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
725 rv = gfs2_glock_nq_atime(&gh);
728 rv = gfs2_ok_for_dio(ip, rw, offset);
730 goto out; /* dio not valid, fall back to buffered i/o */
732 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
733 iov, offset, nr_segs,
734 gfs2_get_block_direct, NULL);
736 gfs2_glock_dq_m(1, &gh);
737 gfs2_holder_uninit(&gh);
742 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
743 * @bh: the buffer we're stuck on
747 static void stuck_releasepage(struct buffer_head *bh)
749 struct inode *inode = bh->b_page->mapping->host;
750 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
751 struct gfs2_bufdata *bd = bh->b_private;
752 struct gfs2_glock *gl;
753 static unsigned limit = 0;
759 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
760 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
761 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
762 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
763 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
770 fs_warn(sdp, "gl = (%u, %llu)\n",
771 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
773 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
774 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
775 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
777 if (gl->gl_ops == &gfs2_inode_glops) {
778 struct gfs2_inode *ip = gl->gl_object;
784 fs_warn(sdp, "ip = %llu %llu\n",
785 (unsigned long long)ip->i_no_formal_ino,
786 (unsigned long long)ip->i_no_addr);
788 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
789 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
790 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
795 * gfs2_releasepage - free the metadata associated with a page
796 * @page: the page that's being released
797 * @gfp_mask: passed from Linux VFS, ignored by us
799 * Call try_to_free_buffers() if the buffers in this page can be
805 int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
807 struct inode *aspace = page->mapping->host;
808 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
809 struct buffer_head *bh, *head;
810 struct gfs2_bufdata *bd;
811 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
813 if (!page_has_buffers(page))
816 head = bh = page_buffers(page);
818 while (atomic_read(&bh->b_count)) {
819 if (!atomic_read(&aspace->i_writecount))
822 if (!(gfp_mask & __GFP_WAIT))
825 if (time_after_eq(jiffies, t)) {
826 stuck_releasepage(bh);
827 /* should we withdraw here? */
834 gfs2_assert_warn(sdp, !buffer_pinned(bh));
835 gfs2_assert_warn(sdp, !buffer_dirty(bh));
840 gfs2_assert_warn(sdp, bd->bd_bh == bh);
841 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
842 gfs2_assert_warn(sdp, !bd->bd_ail);
844 if (!list_empty(&bd->bd_le.le_list))
846 bh->b_private = NULL;
848 gfs2_log_unlock(sdp);
850 kmem_cache_free(gfs2_bufdata_cachep, bd);
852 bh = bh->b_this_page;
853 } while (bh != head);
856 return try_to_free_buffers(page);
859 const struct address_space_operations gfs2_file_aops = {
860 .writepage = gfs2_writepage,
861 .writepages = gfs2_writepages,
862 .readpage = gfs2_readpage,
863 .readpages = gfs2_readpages,
864 .sync_page = block_sync_page,
865 .prepare_write = gfs2_prepare_write,
866 .commit_write = gfs2_commit_write,
867 .set_page_dirty = gfs2_set_page_dirty,
869 .invalidatepage = gfs2_invalidatepage,
870 .releasepage = gfs2_releasepage,
871 .direct_IO = gfs2_direct_IO,