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/gfs2_ondisk.h>
21 #include <linux/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->i_private = 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, &ai->ai_ail2_list);
127 if (!buffer_dirty(bh))
130 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
132 gfs2_log_unlock(sdp);
134 ll_rw_block(WRITE, 1, &bh);
144 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
145 * @sdp: the filesystem
150 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
152 struct gfs2_bufdata *bd, *s;
153 struct buffer_head *bh;
155 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
159 gfs2_assert(sdp, bd->bd_ail == ai);
161 if (buffer_busy(bh)) {
168 if (!buffer_uptodate(bh))
169 gfs2_io_error_bh(sdp, bh);
171 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
174 return list_empty(&ai->ai_ail1_list);
178 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
179 * @sdp: the filesystem
184 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
186 struct list_head *head = &ai->ai_ail2_list;
187 struct gfs2_bufdata *bd;
189 while (!list_empty(head)) {
190 bd = list_entry(head->prev, struct gfs2_bufdata,
192 gfs2_assert(sdp, bd->bd_ail == ai);
194 list_del(&bd->bd_ail_st_list);
195 list_del(&bd->bd_ail_gl_list);
196 atomic_dec(&bd->bd_gl->gl_ail_count);
202 * ail_empty_gl - remove all buffers for a given lock from the AIL
205 * None of the buffers should be dirty, locked, or pinned.
208 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
210 struct gfs2_sbd *sdp = gl->gl_sbd;
212 struct list_head *head = &gl->gl_ail_list;
213 struct gfs2_bufdata *bd;
214 struct buffer_head *bh;
218 blocks = atomic_read(&gl->gl_ail_count);
222 error = gfs2_trans_begin(sdp, 0, blocks);
223 if (gfs2_assert_withdraw(sdp, !error))
227 while (!list_empty(head)) {
228 bd = list_entry(head->next, struct gfs2_bufdata,
231 blkno = bh->b_blocknr;
232 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
235 list_del(&bd->bd_ail_st_list);
236 list_del(&bd->bd_ail_gl_list);
237 atomic_dec(&gl->gl_ail_count);
239 gfs2_log_unlock(sdp);
241 gfs2_trans_add_revoke(sdp, blkno);
245 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
246 gfs2_log_unlock(sdp);
249 gfs2_log_flush(sdp, NULL);
253 * gfs2_meta_inval - Invalidate all buffers associated with a glock
258 void gfs2_meta_inval(struct gfs2_glock *gl)
260 struct gfs2_sbd *sdp = gl->gl_sbd;
261 struct inode *aspace = gl->gl_aspace;
262 struct address_space *mapping = gl->gl_aspace->i_mapping;
264 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
266 atomic_inc(&aspace->i_writecount);
267 truncate_inode_pages(mapping, 0);
268 atomic_dec(&aspace->i_writecount);
270 gfs2_assert_withdraw(sdp, !mapping->nrpages);
274 * gfs2_meta_sync - Sync all buffers associated with a glock
279 void gfs2_meta_sync(struct gfs2_glock *gl)
281 struct address_space *mapping = gl->gl_aspace->i_mapping;
284 filemap_fdatawrite(mapping);
285 error = filemap_fdatawait(mapping);
288 gfs2_io_error(gl->gl_sbd);
292 * getbuf - Get a buffer with a given address space
293 * @sdp: the filesystem
294 * @aspace: the address space
295 * @blkno: the block number (filesystem scope)
296 * @create: 1 if the buffer should be created
298 * Returns: the buffer
301 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
302 u64 blkno, int create)
305 struct buffer_head *bh;
310 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
311 index = blkno >> shift; /* convert block to page */
312 bufnum = blkno - (index << shift); /* block buf index within page */
316 page = grab_cache_page(aspace->i_mapping, index);
322 page = find_lock_page(aspace->i_mapping, index);
327 if (!page_has_buffers(page))
328 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
330 /* Locate header for our buffer within our page */
331 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
335 if (!buffer_mapped(bh))
336 map_bh(bh, sdp->sd_vfs, blkno);
339 mark_page_accessed(page);
340 page_cache_release(page);
345 static void meta_prep_new(struct buffer_head *bh)
347 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
350 clear_buffer_dirty(bh);
351 set_buffer_uptodate(bh);
354 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
358 * gfs2_meta_new - Get a block
359 * @gl: The glock associated with this block
360 * @blkno: The block number
362 * Returns: The buffer
365 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
367 struct buffer_head *bh;
368 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
374 * gfs2_meta_read - Read a block from disk
375 * @gl: The glock covering the block
376 * @blkno: The block number
378 * @bhp: the place where the buffer is returned (NULL on failure)
383 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
384 struct buffer_head **bhp)
386 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
387 if (!buffer_uptodate(*bhp))
388 ll_rw_block(READ, 1, bhp);
389 if (flags & DIO_WAIT) {
390 int error = gfs2_meta_wait(gl->gl_sbd, *bhp);
401 * gfs2_meta_wait - Reread a block from disk
402 * @sdp: the filesystem
403 * @bh: The block to wait for
408 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
410 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
415 if (!buffer_uptodate(bh)) {
416 struct gfs2_trans *tr = current->journal_info;
417 if (tr && tr->tr_touched)
418 gfs2_io_error_bh(sdp, bh);
421 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
428 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
429 * @gl: the glock the buffer belongs to
430 * @bh: The buffer to be attached to
431 * @meta: Flag to indicate whether its metadata or not
434 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
437 struct gfs2_bufdata *bd;
440 lock_page(bh->b_page);
444 unlock_page(bh->b_page);
448 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
449 memset(bd, 0, sizeof(struct gfs2_bufdata));
453 INIT_LIST_HEAD(&bd->bd_list_tr);
455 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
457 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
461 unlock_page(bh->b_page);
465 * gfs2_pin - Pin a buffer in memory
466 * @sdp: the filesystem the buffer belongs to
467 * @bh: The buffer to be pinned
471 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
473 struct gfs2_bufdata *bd = bh->b_private;
475 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
477 if (test_set_buffer_pinned(bh))
478 gfs2_assert_withdraw(sdp, 0);
482 /* If this buffer is in the AIL and it has already been written
483 to in-place disk block, remove it from the AIL. */
486 if (bd->bd_ail && !buffer_in_io(bh))
487 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
488 gfs2_log_unlock(sdp);
490 clear_buffer_dirty(bh);
493 if (!buffer_uptodate(bh))
494 gfs2_io_error_bh(sdp, bh);
500 * gfs2_unpin - Unpin a buffer
501 * @sdp: the filesystem the buffer belongs to
502 * @bh: The buffer to unpin
507 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
510 struct gfs2_bufdata *bd = bh->b_private;
512 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
514 if (!buffer_pinned(bh))
515 gfs2_assert_withdraw(sdp, 0);
517 mark_buffer_dirty(bh);
518 clear_buffer_pinned(bh);
522 list_del(&bd->bd_ail_st_list);
525 struct gfs2_glock *gl = bd->bd_gl;
526 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
527 atomic_inc(&gl->gl_ail_count);
530 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
531 gfs2_log_unlock(sdp);
535 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
536 * @ip: the inode who owns the buffers
537 * @bstart: the first buffer in the run
538 * @blen: the number of buffers in the run
542 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
544 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
545 struct inode *aspace = ip->i_gl->gl_aspace;
546 struct buffer_head *bh;
549 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
551 struct gfs2_bufdata *bd = bh->b_private;
553 if (test_clear_buffer_pinned(bh)) {
554 struct gfs2_trans *tr = current->journal_info;
556 list_del_init(&bd->bd_le.le_list);
557 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
558 sdp->sd_log_num_buf--;
559 gfs2_log_unlock(sdp);
566 u64 blkno = bh->b_blocknr;
568 list_del(&bd->bd_ail_st_list);
569 list_del(&bd->bd_ail_gl_list);
570 atomic_dec(&bd->bd_gl->gl_ail_count);
572 gfs2_log_unlock(sdp);
573 gfs2_trans_add_revoke(sdp, blkno);
575 gfs2_log_unlock(sdp);
579 clear_buffer_dirty(bh);
580 clear_buffer_uptodate(bh);
592 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
593 * @ip: The GFS2 inode
595 * This releases buffers that are in the most-recently-used array of
596 * blocks used for indirect block addressing for this inode.
599 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
601 struct buffer_head **bh_slot;
604 spin_lock(&ip->i_spin);
606 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
607 bh_slot = &ip->i_cache[x];
614 spin_unlock(&ip->i_spin);
618 * gfs2_meta_indirect_buffer - Get a metadata buffer
619 * @ip: The GFS2 inode
620 * @height: The level of this buf in the metadata (indir addr) tree (if any)
621 * @num: The block number (device relative) of the buffer
622 * @new: Non-zero if we may create a new buffer
623 * @bhp: the buffer is returned here
625 * Try to use the gfs2_inode's MRU metadata tree cache.
630 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
631 int new, struct buffer_head **bhp)
633 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
634 struct gfs2_glock *gl = ip->i_gl;
635 struct buffer_head *bh = NULL, **bh_slot = ip->i_cache + height;
638 spin_lock(&ip->i_spin);
639 if (*bh_slot && (*bh_slot)->b_blocknr == num) {
644 spin_unlock(&ip->i_spin);
647 bh = getbuf(gl->gl_sbd, gl->gl_aspace, num, CREATE);
653 if (gfs2_assert_warn(sdp, height))
656 gfs2_trans_add_bh(ip->i_gl, bh, 1);
657 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
658 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
660 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
661 if (!buffer_uptodate(bh)) {
662 ll_rw_block(READ, 1, &bh);
663 if (gfs2_meta_wait(sdp, bh))
666 if (gfs2_metatype_check(sdp, bh, mtype))
671 spin_lock(&ip->i_spin);
676 spin_unlock(&ip->i_spin);
687 * gfs2_meta_ra - start readahead on an extent of a file
688 * @gl: the glock the blocks belong to
689 * @dblock: the starting disk block
690 * @extlen: the number of blocks in the extent
692 * returns: the first buffer in the extent
695 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
697 struct gfs2_sbd *sdp = gl->gl_sbd;
698 struct inode *aspace = gl->gl_aspace;
699 struct buffer_head *first_bh, *bh;
700 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
701 sdp->sd_sb.sb_bsize_shift;
710 first_bh = getbuf(sdp, aspace, dblock, CREATE);
712 if (buffer_uptodate(first_bh))
714 if (!buffer_locked(first_bh))
715 ll_rw_block(READ, 1, &first_bh);
721 bh = getbuf(sdp, aspace, dblock, CREATE);
723 if (!buffer_uptodate(bh) && !buffer_locked(bh))
724 ll_rw_block(READA, 1, &bh);
728 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
732 wait_on_buffer(first_bh);
738 * gfs2_meta_syncfs - sync all the buffers in a filesystem
739 * @sdp: the filesystem
743 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
745 gfs2_log_flush(sdp, NULL);
747 gfs2_ail1_start(sdp, DIO_ALL);
748 if (gfs2_ail1_empty(sdp, DIO_ALL))