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"
35 #define buffer_busy(bh) \
36 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
37 #define buffer_in_io(bh) \
38 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
40 static int aspace_get_block(struct inode *inode, sector_t lblock,
41 struct buffer_head *bh_result, int create)
43 gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
47 static int gfs2_aspace_writepage(struct page *page,
48 struct writeback_control *wbc)
50 return block_write_full_page(page, aspace_get_block, wbc);
54 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
55 * @bh: the buffer we're stuck on
59 static void stuck_releasepage(struct buffer_head *bh)
61 struct inode *inode = bh->b_page->mapping->host;
62 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
63 struct gfs2_bufdata *bd = bh->b_private;
64 struct gfs2_glock *gl;
66 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
67 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
68 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
69 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
70 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
77 fs_warn(sdp, "gl = (%u, %llu)\n",
78 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
80 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
81 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
82 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
84 if (gl->gl_ops == &gfs2_inode_glops) {
85 struct gfs2_inode *ip = gl->gl_object;
91 fs_warn(sdp, "ip = %llu %llu\n",
92 (unsigned long long)ip->i_num.no_formal_ino,
93 (unsigned long long)ip->i_num.no_addr);
94 fs_warn(sdp, "ip->i_count = %d, ip->i_vnode = %s\n",
95 atomic_read(&ip->i_count),
96 (ip->i_vnode) ? "!NULL" : "NULL");
98 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
99 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
100 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
105 * gfs2_aspace_releasepage - free the metadata associated with a page
106 * @page: the page that's being released
107 * @gfp_mask: passed from Linux VFS, ignored by us
109 * Call try_to_free_buffers() if the buffers in this page can be
115 static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask)
117 struct inode *aspace = page->mapping->host;
118 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
119 struct buffer_head *bh, *head;
120 struct gfs2_bufdata *bd;
123 if (!page_has_buffers(page))
126 head = bh = page_buffers(page);
130 while (atomic_read(&bh->b_count)) {
131 if (atomic_read(&aspace->i_writecount)) {
132 if (time_after_eq(jiffies, t +
133 gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
134 stuck_releasepage(bh);
145 gfs2_assert_warn(sdp, !buffer_pinned(bh));
149 gfs2_assert_warn(sdp, bd->bd_bh == bh);
150 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
151 gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
152 gfs2_assert_warn(sdp, !bd->bd_ail);
153 kmem_cache_free(gfs2_bufdata_cachep, bd);
154 bh->b_private = NULL;
157 bh = bh->b_this_page;
162 return try_to_free_buffers(page);
165 static struct address_space_operations aspace_aops = {
166 .writepage = gfs2_aspace_writepage,
167 .releasepage = gfs2_aspace_releasepage,
171 * gfs2_aspace_get - Create and initialize a struct inode structure
172 * @sdp: the filesystem the aspace is in
174 * Right now a struct inode is just a struct inode. Maybe Linux
175 * will supply a more lightweight address space construct (that works)
178 * Make sure pages/buffers in this aspace aren't in high memory.
180 * Returns: the aspace
183 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
185 struct inode *aspace;
187 aspace = new_inode(sdp->sd_vfs);
189 mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL);
190 aspace->i_mapping->a_ops = &aspace_aops;
191 aspace->i_size = ~0ULL;
192 aspace->u.generic_ip = NULL;
193 insert_inode_hash(aspace);
198 void gfs2_aspace_put(struct inode *aspace)
200 remove_inode_hash(aspace);
205 * gfs2_ail1_start_one - Start I/O on a part of the AIL
206 * @sdp: the filesystem
207 * @tr: the part of the AIL
211 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
213 struct gfs2_bufdata *bd, *s;
214 struct buffer_head *bh;
217 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
222 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
226 gfs2_assert(sdp, bd->bd_ail == ai);
228 if (!buffer_busy(bh)) {
229 if (!buffer_uptodate(bh)) {
230 gfs2_log_unlock(sdp);
231 gfs2_io_error_bh(sdp, bh);
234 list_move(&bd->bd_ail_st_list,
239 if (!buffer_dirty(bh))
242 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
244 gfs2_log_unlock(sdp);
246 ll_rw_block(WRITE, 1, &bh);
256 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
257 * @sdp: the filesystem
262 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
264 struct gfs2_bufdata *bd, *s;
265 struct buffer_head *bh;
267 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
271 gfs2_assert(sdp, bd->bd_ail == ai);
273 if (buffer_busy(bh)) {
280 if (!buffer_uptodate(bh))
281 gfs2_io_error_bh(sdp, bh);
283 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
286 return list_empty(&ai->ai_ail1_list);
290 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
291 * @sdp: the filesystem
296 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
298 struct list_head *head = &ai->ai_ail2_list;
299 struct gfs2_bufdata *bd;
301 while (!list_empty(head)) {
302 bd = list_entry(head->prev, struct gfs2_bufdata,
304 gfs2_assert(sdp, bd->bd_ail == ai);
306 list_del(&bd->bd_ail_st_list);
307 list_del(&bd->bd_ail_gl_list);
308 atomic_dec(&bd->bd_gl->gl_ail_count);
314 * ail_empty_gl - remove all buffers for a given lock from the AIL
317 * None of the buffers should be dirty, locked, or pinned.
320 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
322 struct gfs2_sbd *sdp = gl->gl_sbd;
324 struct list_head *head = &gl->gl_ail_list;
325 struct gfs2_bufdata *bd;
326 struct buffer_head *bh;
330 blocks = atomic_read(&gl->gl_ail_count);
334 error = gfs2_trans_begin(sdp, 0, blocks);
335 if (gfs2_assert_withdraw(sdp, !error))
339 while (!list_empty(head)) {
340 bd = list_entry(head->next, struct gfs2_bufdata,
343 blkno = bh->b_blocknr;
344 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
347 list_del(&bd->bd_ail_st_list);
348 list_del(&bd->bd_ail_gl_list);
349 atomic_dec(&gl->gl_ail_count);
351 gfs2_log_unlock(sdp);
353 gfs2_trans_add_revoke(sdp, blkno);
357 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
358 gfs2_log_unlock(sdp);
361 gfs2_log_flush(sdp, NULL);
365 * gfs2_meta_inval - Invalidate all buffers associated with a glock
370 void gfs2_meta_inval(struct gfs2_glock *gl)
372 struct gfs2_sbd *sdp = gl->gl_sbd;
373 struct inode *aspace = gl->gl_aspace;
374 struct address_space *mapping = gl->gl_aspace->i_mapping;
376 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
378 atomic_inc(&aspace->i_writecount);
379 truncate_inode_pages(mapping, 0);
380 atomic_dec(&aspace->i_writecount);
382 gfs2_assert_withdraw(sdp, !mapping->nrpages);
386 * gfs2_meta_sync - Sync all buffers associated with a glock
388 * @flags: DIO_START | DIO_WAIT
392 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
394 struct address_space *mapping = gl->gl_aspace->i_mapping;
397 if (flags & DIO_START)
398 filemap_fdatawrite(mapping);
399 if (!error && (flags & DIO_WAIT))
400 error = filemap_fdatawait(mapping);
403 gfs2_io_error(gl->gl_sbd);
407 * getbuf - Get a buffer with a given address space
408 * @sdp: the filesystem
409 * @aspace: the address space
410 * @blkno: the block number (filesystem scope)
411 * @create: 1 if the buffer should be created
413 * Returns: the buffer
416 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
417 uint64_t blkno, int create)
420 struct buffer_head *bh;
425 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
426 index = blkno >> shift; /* convert block to page */
427 bufnum = blkno - (index << shift); /* block buf index within page */
431 page = grab_cache_page(aspace->i_mapping, index);
437 page = find_lock_page(aspace->i_mapping, index);
442 if (!page_has_buffers(page))
443 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
445 /* Locate header for our buffer within our page */
446 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
450 if (!buffer_mapped(bh))
451 map_bh(bh, sdp->sd_vfs, blkno);
454 mark_page_accessed(page);
455 page_cache_release(page);
460 static void meta_prep_new(struct buffer_head *bh)
462 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
465 clear_buffer_dirty(bh);
466 set_buffer_uptodate(bh);
469 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
473 * gfs2_meta_new - Get a block
474 * @gl: The glock associated with this block
475 * @blkno: The block number
477 * Returns: The buffer
480 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
482 struct buffer_head *bh;
483 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
489 * gfs2_meta_read - Read a block from disk
490 * @gl: The glock covering the block
491 * @blkno: The block number
492 * @flags: flags to gfs2_dreread()
493 * @bhp: the place where the buffer is returned (NULL on failure)
498 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
499 struct buffer_head **bhp)
503 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
504 error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
512 * gfs2_meta_reread - Reread a block from disk
513 * @sdp: the filesystem
514 * @bh: The block to read
515 * @flags: Flags that control the read
520 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
522 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
525 if (flags & DIO_FORCE)
526 clear_buffer_uptodate(bh);
528 if ((flags & DIO_START) && !buffer_uptodate(bh))
529 ll_rw_block(READ, 1, &bh);
531 if (flags & DIO_WAIT) {
534 if (!buffer_uptodate(bh)) {
535 struct gfs2_trans *tr = current->journal_info;
536 if (tr && tr->tr_touched)
537 gfs2_io_error_bh(sdp, bh);
540 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
548 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
549 * @gl: the glock the buffer belongs to
550 * @bh: The buffer to be attached to
551 * @meta: Flag to indicate whether its metadata or not
554 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
557 struct gfs2_bufdata *bd;
560 lock_page(bh->b_page);
564 unlock_page(bh->b_page);
568 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
569 memset(bd, 0, sizeof(struct gfs2_bufdata));
574 INIT_LIST_HEAD(&bd->bd_list_tr);
576 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
578 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
584 unlock_page(bh->b_page);
588 * gfs2_pin - Pin a buffer in memory
589 * @sdp: the filesystem the buffer belongs to
590 * @bh: The buffer to be pinned
594 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
596 struct gfs2_bufdata *bd = bh->b_private;
598 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
600 if (test_set_buffer_pinned(bh))
601 gfs2_assert_withdraw(sdp, 0);
605 /* If this buffer is in the AIL and it has already been written
606 to in-place disk block, remove it from the AIL. */
609 if (bd->bd_ail && !buffer_in_io(bh))
610 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
611 gfs2_log_unlock(sdp);
613 clear_buffer_dirty(bh);
616 if (!buffer_uptodate(bh))
617 gfs2_io_error_bh(sdp, bh);
623 * gfs2_unpin - Unpin a buffer
624 * @sdp: the filesystem the buffer belongs to
625 * @bh: The buffer to unpin
630 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
633 struct gfs2_bufdata *bd = bh->b_private;
635 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
637 if (!buffer_pinned(bh))
638 gfs2_assert_withdraw(sdp, 0);
640 mark_buffer_dirty(bh);
641 clear_buffer_pinned(bh);
645 list_del(&bd->bd_ail_st_list);
648 struct gfs2_glock *gl = bd->bd_gl;
649 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
650 atomic_inc(&gl->gl_ail_count);
653 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
654 gfs2_log_unlock(sdp);
658 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
659 * @ip: the inode who owns the buffers
660 * @bstart: the first buffer in the run
661 * @blen: the number of buffers in the run
665 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
667 struct gfs2_sbd *sdp = ip->i_sbd;
668 struct inode *aspace = ip->i_gl->gl_aspace;
669 struct buffer_head *bh;
672 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
674 struct gfs2_bufdata *bd = bh->b_private;
676 if (test_clear_buffer_pinned(bh)) {
677 struct gfs2_trans *tr = current->journal_info;
679 list_del_init(&bd->bd_le.le_list);
680 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
681 sdp->sd_log_num_buf--;
682 gfs2_log_unlock(sdp);
689 uint64_t blkno = bh->b_blocknr;
691 list_del(&bd->bd_ail_st_list);
692 list_del(&bd->bd_ail_gl_list);
693 atomic_dec(&bd->bd_gl->gl_ail_count);
695 gfs2_log_unlock(sdp);
696 gfs2_trans_add_revoke(sdp, blkno);
698 gfs2_log_unlock(sdp);
702 clear_buffer_dirty(bh);
703 clear_buffer_uptodate(bh);
715 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
716 * @ip: The GFS2 inode
718 * This releases buffers that are in the most-recently-used array of
719 * blocks used for indirect block addressing for this inode.
722 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
724 struct buffer_head **bh_slot;
727 spin_lock(&ip->i_spin);
729 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
730 bh_slot = &ip->i_cache[x];
737 spin_unlock(&ip->i_spin);
741 * gfs2_meta_indirect_buffer - Get a metadata buffer
742 * @ip: The GFS2 inode
743 * @height: The level of this buf in the metadata (indir addr) tree (if any)
744 * @num: The block number (device relative) of the buffer
745 * @new: Non-zero if we may create a new buffer
746 * @bhp: the buffer is returned here
748 * Try to use the gfs2_inode's MRU metadata tree cache.
753 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
754 int new, struct buffer_head **bhp)
756 struct buffer_head *bh, **bh_slot = ip->i_cache + height;
759 spin_lock(&ip->i_spin);
762 if (bh->b_blocknr == num)
767 spin_unlock(&ip->i_spin);
773 error = gfs2_meta_reread(ip->i_sbd, bh,
774 DIO_START | DIO_WAIT);
782 bh = gfs2_meta_new(ip->i_gl, num);
784 error = gfs2_meta_read(ip->i_gl, num,
785 DIO_START | DIO_WAIT, &bh);
790 spin_lock(&ip->i_spin);
791 if (*bh_slot != bh) {
796 spin_unlock(&ip->i_spin);
800 if (gfs2_assert_warn(ip->i_sbd, height)) {
804 gfs2_trans_add_bh(ip->i_gl, bh, 1);
805 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
806 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
808 } else if (gfs2_metatype_check(ip->i_sbd, bh,
809 (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
820 * gfs2_meta_ra - start readahead on an extent of a file
821 * @gl: the glock the blocks belong to
822 * @dblock: the starting disk block
823 * @extlen: the number of blocks in the extent
827 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
829 struct gfs2_sbd *sdp = gl->gl_sbd;
830 struct inode *aspace = gl->gl_aspace;
831 struct buffer_head *first_bh, *bh;
832 uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
833 sdp->sd_sb.sb_bsize_shift;
836 if (!extlen || !max_ra)
841 first_bh = getbuf(sdp, aspace, dblock, CREATE);
843 if (buffer_uptodate(first_bh))
845 if (!buffer_locked(first_bh)) {
846 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
855 bh = getbuf(sdp, aspace, dblock, CREATE);
857 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
858 error = gfs2_meta_reread(sdp, bh, DIO_START);
868 if (buffer_uptodate(first_bh))
877 * gfs2_meta_syncfs - sync all the buffers in a filesystem
878 * @sdp: the filesystem
882 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
884 gfs2_log_flush(sdp, NULL);
886 gfs2_ail1_start(sdp, DIO_ALL);
887 if (gfs2_ail1_empty(sdp, DIO_ALL))