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 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>
16 #include <linux/pagemap.h>
17 #include <linux/writeback.h>
18 #include <linux/swap.h>
19 #include <linux/delay.h>
20 #include <linux/bio.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/lm_interface.h>
35 #include "ops_address.h"
37 #define buffer_busy(bh) \
38 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
39 #define buffer_in_io(bh) \
40 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
42 static int aspace_get_block(struct inode *inode, sector_t lblock,
43 struct buffer_head *bh_result, int create)
45 gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
49 static int gfs2_aspace_writepage(struct page *page,
50 struct writeback_control *wbc)
52 return block_write_full_page(page, aspace_get_block, wbc);
55 static const struct address_space_operations aspace_aops = {
56 .writepage = gfs2_aspace_writepage,
57 .releasepage = gfs2_releasepage,
61 * gfs2_aspace_get - Create and initialize a struct inode structure
62 * @sdp: the filesystem the aspace is in
64 * Right now a struct inode is just a struct inode. Maybe Linux
65 * will supply a more lightweight address space construct (that works)
68 * Make sure pages/buffers in this aspace aren't in high memory.
73 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
77 aspace = new_inode(sdp->sd_vfs);
79 mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS);
80 aspace->i_mapping->a_ops = &aspace_aops;
81 aspace->i_size = ~0ULL;
82 aspace->i_private = NULL;
83 insert_inode_hash(aspace);
88 void gfs2_aspace_put(struct inode *aspace)
90 remove_inode_hash(aspace);
95 * gfs2_ail1_start_one - Start I/O on a part of the AIL
96 * @sdp: the filesystem
97 * @tr: the part of the AIL
101 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
103 struct gfs2_bufdata *bd, *s;
104 struct buffer_head *bh;
107 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
112 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
116 gfs2_assert(sdp, bd->bd_ail == ai);
118 if (!buffer_busy(bh)) {
119 if (!buffer_uptodate(bh)) {
120 gfs2_log_unlock(sdp);
121 gfs2_io_error_bh(sdp, bh);
124 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_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
280 void gfs2_meta_sync(struct gfs2_glock *gl)
282 struct address_space *mapping = gl->gl_aspace->i_mapping;
285 filemap_fdatawrite(mapping);
286 error = filemap_fdatawait(mapping);
289 gfs2_io_error(gl->gl_sbd);
293 * getbuf - Get a buffer with a given address space
294 * @sdp: the filesystem
295 * @aspace: the address space
296 * @blkno: the block number (filesystem scope)
297 * @create: 1 if the buffer should be created
299 * Returns: the buffer
302 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
303 u64 blkno, int create)
306 struct buffer_head *bh;
311 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
312 index = blkno >> shift; /* convert block to page */
313 bufnum = blkno - (index << shift); /* block buf index within page */
317 page = grab_cache_page(aspace->i_mapping, index);
323 page = find_lock_page(aspace->i_mapping, index);
328 if (!page_has_buffers(page))
329 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
331 /* Locate header for our buffer within our page */
332 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
336 if (!buffer_mapped(bh))
337 map_bh(bh, sdp->sd_vfs, blkno);
340 mark_page_accessed(page);
341 page_cache_release(page);
346 static void meta_prep_new(struct buffer_head *bh)
348 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
351 clear_buffer_dirty(bh);
352 set_buffer_uptodate(bh);
355 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
359 * gfs2_meta_new - Get a block
360 * @gl: The glock associated with this block
361 * @blkno: The block number
363 * Returns: The buffer
366 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
368 struct buffer_head *bh;
369 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
375 * gfs2_meta_read - Read a block from disk
376 * @gl: The glock covering the block
377 * @blkno: The block number
379 * @bhp: the place where the buffer is returned (NULL on failure)
384 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
385 struct buffer_head **bhp)
387 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
388 if (!buffer_uptodate(*bhp))
389 ll_rw_block(READ_META, 1, bhp);
390 if (flags & DIO_WAIT) {
391 int error = gfs2_meta_wait(gl->gl_sbd, *bhp);
402 * gfs2_meta_wait - Reread a block from disk
403 * @sdp: the filesystem
404 * @bh: The block to wait for
409 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
411 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
416 if (!buffer_uptodate(bh)) {
417 struct gfs2_trans *tr = current->journal_info;
418 if (tr && tr->tr_touched)
419 gfs2_io_error_bh(sdp, bh);
422 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
429 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
430 * @gl: the glock the buffer belongs to
431 * @bh: The buffer to be attached to
432 * @meta: Flag to indicate whether its metadata or not
435 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
438 struct gfs2_bufdata *bd;
441 lock_page(bh->b_page);
445 unlock_page(bh->b_page);
449 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
450 memset(bd, 0, sizeof(struct gfs2_bufdata));
454 INIT_LIST_HEAD(&bd->bd_list_tr);
456 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
458 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
462 unlock_page(bh->b_page);
466 * gfs2_pin - Pin a buffer in memory
467 * @sdp: the filesystem the buffer belongs to
468 * @bh: The buffer to be pinned
472 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
474 struct gfs2_bufdata *bd = bh->b_private;
476 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
478 if (test_set_buffer_pinned(bh))
479 gfs2_assert_withdraw(sdp, 0);
483 /* If this buffer is in the AIL and it has already been written
484 to in-place disk block, remove it from the AIL. */
487 if (bd->bd_ail && !buffer_in_io(bh))
488 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
489 gfs2_log_unlock(sdp);
491 clear_buffer_dirty(bh);
494 if (!buffer_uptodate(bh))
495 gfs2_io_error_bh(sdp, bh);
501 * gfs2_unpin - Unpin a buffer
502 * @sdp: the filesystem the buffer belongs to
503 * @bh: The buffer to unpin
508 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
511 struct gfs2_bufdata *bd = bh->b_private;
513 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
515 if (!buffer_pinned(bh))
516 gfs2_assert_withdraw(sdp, 0);
518 mark_buffer_dirty(bh);
519 clear_buffer_pinned(bh);
523 list_del(&bd->bd_ail_st_list);
526 struct gfs2_glock *gl = bd->bd_gl;
527 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
528 atomic_inc(&gl->gl_ail_count);
531 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
532 gfs2_log_unlock(sdp);
536 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
537 * @ip: the inode who owns the buffers
538 * @bstart: the first buffer in the run
539 * @blen: the number of buffers in the run
543 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
545 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
546 struct inode *aspace = ip->i_gl->gl_aspace;
547 struct buffer_head *bh;
550 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
552 struct gfs2_bufdata *bd = bh->b_private;
554 if (test_clear_buffer_pinned(bh)) {
555 struct gfs2_trans *tr = current->journal_info;
557 list_del_init(&bd->bd_le.le_list);
558 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
559 sdp->sd_log_num_buf--;
560 gfs2_log_unlock(sdp);
567 u64 blkno = bh->b_blocknr;
569 list_del(&bd->bd_ail_st_list);
570 list_del(&bd->bd_ail_gl_list);
571 atomic_dec(&bd->bd_gl->gl_ail_count);
573 gfs2_log_unlock(sdp);
574 gfs2_trans_add_revoke(sdp, blkno);
576 gfs2_log_unlock(sdp);
580 clear_buffer_dirty(bh);
581 clear_buffer_uptodate(bh);
593 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
594 * @ip: The GFS2 inode
596 * This releases buffers that are in the most-recently-used array of
597 * blocks used for indirect block addressing for this inode.
600 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
602 struct buffer_head **bh_slot;
605 spin_lock(&ip->i_spin);
607 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
608 bh_slot = &ip->i_cache[x];
615 spin_unlock(&ip->i_spin);
619 * gfs2_meta_indirect_buffer - Get a metadata buffer
620 * @ip: The GFS2 inode
621 * @height: The level of this buf in the metadata (indir addr) tree (if any)
622 * @num: The block number (device relative) of the buffer
623 * @new: Non-zero if we may create a new buffer
624 * @bhp: the buffer is returned here
626 * Try to use the gfs2_inode's MRU metadata tree cache.
631 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
632 int new, struct buffer_head **bhp)
634 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
635 struct gfs2_glock *gl = ip->i_gl;
636 struct buffer_head *bh = NULL, **bh_slot = ip->i_cache + height;
639 spin_lock(&ip->i_spin);
640 if (*bh_slot && (*bh_slot)->b_blocknr == num) {
645 spin_unlock(&ip->i_spin);
648 bh = getbuf(gl->gl_sbd, gl->gl_aspace, num, CREATE);
654 if (gfs2_assert_warn(sdp, height))
657 gfs2_trans_add_bh(ip->i_gl, bh, 1);
658 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
659 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
661 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
662 if (!buffer_uptodate(bh)) {
663 ll_rw_block(READ_META, 1, &bh);
664 if (gfs2_meta_wait(sdp, bh))
667 if (gfs2_metatype_check(sdp, bh, mtype))
672 spin_lock(&ip->i_spin);
677 spin_unlock(&ip->i_spin);
688 * gfs2_meta_ra - start readahead on an extent of a file
689 * @gl: the glock the blocks belong to
690 * @dblock: the starting disk block
691 * @extlen: the number of blocks in the extent
693 * returns: the first buffer in the extent
696 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
698 struct gfs2_sbd *sdp = gl->gl_sbd;
699 struct inode *aspace = gl->gl_aspace;
700 struct buffer_head *first_bh, *bh;
701 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
702 sdp->sd_sb.sb_bsize_shift;
711 first_bh = getbuf(sdp, aspace, dblock, CREATE);
713 if (buffer_uptodate(first_bh))
715 if (!buffer_locked(first_bh))
716 ll_rw_block(READ_META, 1, &first_bh);
722 bh = getbuf(sdp, aspace, dblock, CREATE);
724 if (!buffer_uptodate(bh) && !buffer_locked(bh))
725 ll_rw_block(READA, 1, &bh);
729 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
733 wait_on_buffer(first_bh);
739 * gfs2_meta_syncfs - sync all the buffers in a filesystem
740 * @sdp: the filesystem
744 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
746 gfs2_log_flush(sdp, NULL);
748 gfs2_ail1_start(sdp, DIO_ALL);
749 if (gfs2_ail1_empty(sdp, DIO_ALL))