2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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 v.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>
16 #include <linux/pagemap.h>
17 #include <linux/writeback.h>
18 #include <linux/swap.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
23 #include "lm_interface.h"
34 #include "ops_address.h"
36 #define buffer_busy(bh) \
37 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
38 #define buffer_in_io(bh) \
39 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
41 static int aspace_get_block(struct inode *inode, sector_t lblock,
42 struct buffer_head *bh_result, int create)
44 gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
48 static int gfs2_aspace_writepage(struct page *page,
49 struct writeback_control *wbc)
51 return block_write_full_page(page, aspace_get_block, wbc);
54 static const struct address_space_operations aspace_aops = {
55 .writepage = gfs2_aspace_writepage,
56 .releasepage = gfs2_releasepage,
60 * gfs2_aspace_get - Create and initialize a struct inode structure
61 * @sdp: the filesystem the aspace is in
63 * Right now a struct inode is just a struct inode. Maybe Linux
64 * will supply a more lightweight address space construct (that works)
67 * Make sure pages/buffers in this aspace aren't in high memory.
72 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
76 aspace = new_inode(sdp->sd_vfs);
78 mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS);
79 aspace->i_mapping->a_ops = &aspace_aops;
80 aspace->i_size = ~0ULL;
81 aspace->u.generic_ip = NULL;
82 insert_inode_hash(aspace);
87 void gfs2_aspace_put(struct inode *aspace)
89 remove_inode_hash(aspace);
94 * gfs2_ail1_start_one - Start I/O on a part of the AIL
95 * @sdp: the filesystem
96 * @tr: the part of the AIL
100 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
102 struct gfs2_bufdata *bd, *s;
103 struct buffer_head *bh;
106 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
111 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
115 gfs2_assert(sdp, bd->bd_ail == ai);
117 if (!buffer_busy(bh)) {
118 if (!buffer_uptodate(bh)) {
119 gfs2_log_unlock(sdp);
120 gfs2_io_error_bh(sdp, bh);
123 list_move(&bd->bd_ail_st_list,
128 if (!buffer_dirty(bh))
131 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
133 gfs2_log_unlock(sdp);
135 ll_rw_block(WRITE, 1, &bh);
145 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
146 * @sdp: the filesystem
151 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
153 struct gfs2_bufdata *bd, *s;
154 struct buffer_head *bh;
156 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
160 gfs2_assert(sdp, bd->bd_ail == ai);
162 if (buffer_busy(bh)) {
169 if (!buffer_uptodate(bh))
170 gfs2_io_error_bh(sdp, bh);
172 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
175 return list_empty(&ai->ai_ail1_list);
179 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
180 * @sdp: the filesystem
185 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
187 struct list_head *head = &ai->ai_ail2_list;
188 struct gfs2_bufdata *bd;
190 while (!list_empty(head)) {
191 bd = list_entry(head->prev, struct gfs2_bufdata,
193 gfs2_assert(sdp, bd->bd_ail == ai);
195 list_del(&bd->bd_ail_st_list);
196 list_del(&bd->bd_ail_gl_list);
197 atomic_dec(&bd->bd_gl->gl_ail_count);
203 * ail_empty_gl - remove all buffers for a given lock from the AIL
206 * None of the buffers should be dirty, locked, or pinned.
209 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
211 struct gfs2_sbd *sdp = gl->gl_sbd;
213 struct list_head *head = &gl->gl_ail_list;
214 struct gfs2_bufdata *bd;
215 struct buffer_head *bh;
219 blocks = atomic_read(&gl->gl_ail_count);
223 error = gfs2_trans_begin(sdp, 0, blocks);
224 if (gfs2_assert_withdraw(sdp, !error))
228 while (!list_empty(head)) {
229 bd = list_entry(head->next, struct gfs2_bufdata,
232 blkno = bh->b_blocknr;
233 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
236 list_del(&bd->bd_ail_st_list);
237 list_del(&bd->bd_ail_gl_list);
238 atomic_dec(&gl->gl_ail_count);
240 gfs2_log_unlock(sdp);
242 gfs2_trans_add_revoke(sdp, blkno);
246 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
247 gfs2_log_unlock(sdp);
250 gfs2_log_flush(sdp, NULL);
254 * gfs2_meta_inval - Invalidate all buffers associated with a glock
259 void gfs2_meta_inval(struct gfs2_glock *gl)
261 struct gfs2_sbd *sdp = gl->gl_sbd;
262 struct inode *aspace = gl->gl_aspace;
263 struct address_space *mapping = gl->gl_aspace->i_mapping;
265 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
267 atomic_inc(&aspace->i_writecount);
268 truncate_inode_pages(mapping, 0);
269 atomic_dec(&aspace->i_writecount);
271 gfs2_assert_withdraw(sdp, !mapping->nrpages);
275 * gfs2_meta_sync - Sync all buffers associated with a glock
277 * @flags: DIO_START | DIO_WAIT
281 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
283 struct address_space *mapping = gl->gl_aspace->i_mapping;
286 if (flags & DIO_START)
287 filemap_fdatawrite(mapping);
288 if (!error && (flags & DIO_WAIT))
289 error = filemap_fdatawait(mapping);
292 gfs2_io_error(gl->gl_sbd);
296 * getbuf - Get a buffer with a given address space
297 * @sdp: the filesystem
298 * @aspace: the address space
299 * @blkno: the block number (filesystem scope)
300 * @create: 1 if the buffer should be created
302 * Returns: the buffer
305 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
306 uint64_t blkno, int create)
309 struct buffer_head *bh;
314 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
315 index = blkno >> shift; /* convert block to page */
316 bufnum = blkno - (index << shift); /* block buf index within page */
320 page = grab_cache_page(aspace->i_mapping, index);
326 page = find_lock_page(aspace->i_mapping, index);
331 if (!page_has_buffers(page))
332 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
334 /* Locate header for our buffer within our page */
335 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
339 if (!buffer_mapped(bh))
340 map_bh(bh, sdp->sd_vfs, blkno);
343 mark_page_accessed(page);
344 page_cache_release(page);
349 static void meta_prep_new(struct buffer_head *bh)
351 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
354 clear_buffer_dirty(bh);
355 set_buffer_uptodate(bh);
358 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
362 * gfs2_meta_new - Get a block
363 * @gl: The glock associated with this block
364 * @blkno: The block number
366 * Returns: The buffer
369 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
371 struct buffer_head *bh;
372 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
378 * gfs2_meta_read - Read a block from disk
379 * @gl: The glock covering the block
380 * @blkno: The block number
381 * @flags: flags to gfs2_dreread()
382 * @bhp: the place where the buffer is returned (NULL on failure)
387 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
388 struct buffer_head **bhp)
392 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
393 error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
401 * gfs2_meta_reread - Reread a block from disk
402 * @sdp: the filesystem
403 * @bh: The block to read
404 * @flags: Flags that control the read
409 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
411 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
414 if (flags & DIO_FORCE)
415 clear_buffer_uptodate(bh);
417 if ((flags & DIO_START) && !buffer_uptodate(bh))
418 ll_rw_block(READ, 1, &bh);
420 if (flags & DIO_WAIT) {
423 if (!buffer_uptodate(bh)) {
424 struct gfs2_trans *tr = current->journal_info;
425 if (tr && tr->tr_touched)
426 gfs2_io_error_bh(sdp, bh);
429 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
437 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
438 * @gl: the glock the buffer belongs to
439 * @bh: The buffer to be attached to
440 * @meta: Flag to indicate whether its metadata or not
443 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
446 struct gfs2_bufdata *bd;
449 lock_page(bh->b_page);
453 unlock_page(bh->b_page);
457 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
458 memset(bd, 0, sizeof(struct gfs2_bufdata));
462 INIT_LIST_HEAD(&bd->bd_list_tr);
464 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
466 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
472 unlock_page(bh->b_page);
476 * gfs2_pin - Pin a buffer in memory
477 * @sdp: the filesystem the buffer belongs to
478 * @bh: The buffer to be pinned
482 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
484 struct gfs2_bufdata *bd = bh->b_private;
486 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
488 if (test_set_buffer_pinned(bh))
489 gfs2_assert_withdraw(sdp, 0);
493 /* If this buffer is in the AIL and it has already been written
494 to in-place disk block, remove it from the AIL. */
497 if (bd->bd_ail && !buffer_in_io(bh))
498 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
499 gfs2_log_unlock(sdp);
501 clear_buffer_dirty(bh);
504 if (!buffer_uptodate(bh))
505 gfs2_io_error_bh(sdp, bh);
511 * gfs2_unpin - Unpin a buffer
512 * @sdp: the filesystem the buffer belongs to
513 * @bh: The buffer to unpin
518 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
521 struct gfs2_bufdata *bd = bh->b_private;
523 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
525 if (!buffer_pinned(bh))
526 gfs2_assert_withdraw(sdp, 0);
528 mark_buffer_dirty(bh);
529 clear_buffer_pinned(bh);
533 list_del(&bd->bd_ail_st_list);
536 struct gfs2_glock *gl = bd->bd_gl;
537 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
538 atomic_inc(&gl->gl_ail_count);
541 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
542 gfs2_log_unlock(sdp);
546 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
547 * @ip: the inode who owns the buffers
548 * @bstart: the first buffer in the run
549 * @blen: the number of buffers in the run
553 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
555 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
556 struct inode *aspace = ip->i_gl->gl_aspace;
557 struct buffer_head *bh;
560 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
562 struct gfs2_bufdata *bd = bh->b_private;
564 if (test_clear_buffer_pinned(bh)) {
565 struct gfs2_trans *tr = current->journal_info;
567 list_del_init(&bd->bd_le.le_list);
568 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
569 sdp->sd_log_num_buf--;
570 gfs2_log_unlock(sdp);
577 uint64_t blkno = bh->b_blocknr;
579 list_del(&bd->bd_ail_st_list);
580 list_del(&bd->bd_ail_gl_list);
581 atomic_dec(&bd->bd_gl->gl_ail_count);
583 gfs2_log_unlock(sdp);
584 gfs2_trans_add_revoke(sdp, blkno);
586 gfs2_log_unlock(sdp);
590 clear_buffer_dirty(bh);
591 clear_buffer_uptodate(bh);
603 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
604 * @ip: The GFS2 inode
606 * This releases buffers that are in the most-recently-used array of
607 * blocks used for indirect block addressing for this inode.
610 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
612 struct buffer_head **bh_slot;
615 spin_lock(&ip->i_spin);
617 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
618 bh_slot = &ip->i_cache[x];
625 spin_unlock(&ip->i_spin);
629 * gfs2_meta_indirect_buffer - Get a metadata buffer
630 * @ip: The GFS2 inode
631 * @height: The level of this buf in the metadata (indir addr) tree (if any)
632 * @num: The block number (device relative) of the buffer
633 * @new: Non-zero if we may create a new buffer
634 * @bhp: the buffer is returned here
636 * Try to use the gfs2_inode's MRU metadata tree cache.
641 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
642 int new, struct buffer_head **bhp)
644 struct buffer_head *bh, **bh_slot = ip->i_cache + height;
647 spin_lock(&ip->i_spin);
650 if (bh->b_blocknr == num)
655 spin_unlock(&ip->i_spin);
661 error = gfs2_meta_reread(GFS2_SB(&ip->i_inode), bh,
662 DIO_START | DIO_WAIT);
670 bh = gfs2_meta_new(ip->i_gl, num);
672 error = gfs2_meta_read(ip->i_gl, num,
673 DIO_START | DIO_WAIT, &bh);
678 spin_lock(&ip->i_spin);
679 if (*bh_slot != bh) {
684 spin_unlock(&ip->i_spin);
688 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), height)) {
692 gfs2_trans_add_bh(ip->i_gl, bh, 1);
693 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
694 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
696 } else if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh,
697 (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
708 * gfs2_meta_ra - start readahead on an extent of a file
709 * @gl: the glock the blocks belong to
710 * @dblock: the starting disk block
711 * @extlen: the number of blocks in the extent
715 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
717 struct gfs2_sbd *sdp = gl->gl_sbd;
718 struct inode *aspace = gl->gl_aspace;
719 struct buffer_head *first_bh, *bh;
720 uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
721 sdp->sd_sb.sb_bsize_shift;
724 if (!extlen || !max_ra)
729 first_bh = getbuf(sdp, aspace, dblock, CREATE);
731 if (buffer_uptodate(first_bh))
733 if (!buffer_locked(first_bh)) {
734 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
743 bh = getbuf(sdp, aspace, dblock, CREATE);
745 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
746 error = gfs2_meta_reread(sdp, bh, DIO_START);
756 if (buffer_uptodate(first_bh))
765 * gfs2_meta_syncfs - sync all the buffers in a filesystem
766 * @sdp: the filesystem
770 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
772 gfs2_log_flush(sdp, NULL);
774 gfs2_ail1_start(sdp, DIO_ALL);
775 if (gfs2_ail1_empty(sdp, DIO_ALL))