2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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>
21 #include <asm/semaphore.h>
24 #include "lm_interface.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);
55 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
56 * @bh: the buffer we're stuck on
60 static void stuck_releasepage(struct buffer_head *bh)
62 struct inode *inode = bh->b_page->mapping->host;
63 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
64 struct gfs2_bufdata *bd = bh->b_private;
65 struct gfs2_glock *gl;
67 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
68 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
69 (uint64_t)bh->b_blocknr, atomic_read(&bh->b_count));
70 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
71 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
78 fs_warn(sdp, "gl = (%u, %llu)\n",
79 gl->gl_name.ln_type, gl->gl_name.ln_number);
81 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
82 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
83 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
85 if (gl->gl_ops == &gfs2_inode_glops) {
86 struct gfs2_inode *ip = gl->gl_object;
92 fs_warn(sdp, "ip = %llu %llu\n",
93 ip->i_num.no_formal_ino, 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;
220 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
224 gfs2_assert(sdp, bd->bd_ail == ai);
226 if (!buffer_busy(bh)) {
227 if (!buffer_uptodate(bh))
228 gfs2_io_error_bh(sdp, bh);
229 list_move(&bd->bd_ail_st_list,
234 if (!buffer_dirty(bh))
237 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
239 gfs2_log_unlock(sdp);
241 ll_rw_block(WRITE, 1, &bh);
251 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
252 * @sdp: the filesystem
257 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
259 struct gfs2_bufdata *bd, *s;
260 struct buffer_head *bh;
262 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
266 gfs2_assert(sdp, bd->bd_ail == ai);
268 if (buffer_busy(bh)) {
275 if (!buffer_uptodate(bh))
276 gfs2_io_error_bh(sdp, bh);
278 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
281 return list_empty(&ai->ai_ail1_list);
285 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
286 * @sdp: the filesystem
291 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
293 struct list_head *head = &ai->ai_ail2_list;
294 struct gfs2_bufdata *bd;
296 while (!list_empty(head)) {
297 bd = list_entry(head->prev, struct gfs2_bufdata,
299 gfs2_assert(sdp, bd->bd_ail == ai);
301 list_del(&bd->bd_ail_st_list);
302 list_del(&bd->bd_ail_gl_list);
303 atomic_dec(&bd->bd_gl->gl_ail_count);
309 * ail_empty_gl - remove all buffers for a given lock from the AIL
312 * None of the buffers should be dirty, locked, or pinned.
315 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
317 struct gfs2_sbd *sdp = gl->gl_sbd;
319 struct list_head *head = &gl->gl_ail_list;
320 struct gfs2_bufdata *bd;
321 struct buffer_head *bh;
325 blocks = atomic_read(&gl->gl_ail_count);
329 error = gfs2_trans_begin(sdp, 0, blocks);
330 if (gfs2_assert_withdraw(sdp, !error))
334 while (!list_empty(head)) {
335 bd = list_entry(head->next, struct gfs2_bufdata,
338 blkno = bh->b_blocknr;
339 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
342 list_del(&bd->bd_ail_st_list);
343 list_del(&bd->bd_ail_gl_list);
344 atomic_dec(&gl->gl_ail_count);
346 gfs2_log_unlock(sdp);
348 gfs2_trans_add_revoke(sdp, blkno);
352 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
353 gfs2_log_unlock(sdp);
356 gfs2_log_flush(sdp, NULL);
360 * gfs2_meta_inval - Invalidate all buffers associated with a glock
365 void gfs2_meta_inval(struct gfs2_glock *gl)
367 struct gfs2_sbd *sdp = gl->gl_sbd;
368 struct inode *aspace = gl->gl_aspace;
369 struct address_space *mapping = gl->gl_aspace->i_mapping;
371 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
373 atomic_inc(&aspace->i_writecount);
374 truncate_inode_pages(mapping, 0);
375 atomic_dec(&aspace->i_writecount);
377 gfs2_assert_withdraw(sdp, !mapping->nrpages);
381 * gfs2_meta_sync - Sync all buffers associated with a glock
383 * @flags: DIO_START | DIO_WAIT
387 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
389 struct address_space *mapping = gl->gl_aspace->i_mapping;
392 if (flags & DIO_START)
393 filemap_fdatawrite(mapping);
394 if (!error && (flags & DIO_WAIT))
395 error = filemap_fdatawait(mapping);
398 gfs2_io_error(gl->gl_sbd);
402 * getbuf - Get a buffer with a given address space
403 * @sdp: the filesystem
404 * @aspace: the address space
405 * @blkno: the block number (filesystem scope)
406 * @create: 1 if the buffer should be created
408 * Returns: the buffer
411 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
412 uint64_t blkno, int create)
415 struct buffer_head *bh;
420 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
421 index = blkno >> shift; /* convert block to page */
422 bufnum = blkno - (index << shift); /* block buf index within page */
426 page = grab_cache_page(aspace->i_mapping, index);
432 page = find_lock_page(aspace->i_mapping, index);
437 if (!page_has_buffers(page))
438 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
440 /* Locate header for our buffer within our page */
441 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
445 if (!buffer_mapped(bh))
446 map_bh(bh, sdp->sd_vfs, blkno);
449 mark_page_accessed(page);
450 page_cache_release(page);
455 static void meta_prep_new(struct buffer_head *bh)
457 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
460 clear_buffer_dirty(bh);
461 set_buffer_uptodate(bh);
464 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
468 * gfs2_meta_new - Get a block
469 * @gl: The glock associated with this block
470 * @blkno: The block number
472 * Returns: The buffer
475 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
477 struct buffer_head *bh;
478 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
484 * gfs2_meta_read - Read a block from disk
485 * @gl: The glock covering the block
486 * @blkno: The block number
487 * @flags: flags to gfs2_dreread()
488 * @bhp: the place where the buffer is returned (NULL on failure)
493 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
494 struct buffer_head **bhp)
498 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
499 error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
507 * gfs2_meta_reread - Reread a block from disk
508 * @sdp: the filesystem
509 * @bh: The block to read
510 * @flags: Flags that control the read
515 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
517 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
520 if (flags & DIO_FORCE)
521 clear_buffer_uptodate(bh);
523 if ((flags & DIO_START) && !buffer_uptodate(bh))
524 ll_rw_block(READ, 1, &bh);
526 if (flags & DIO_WAIT) {
529 if (!buffer_uptodate(bh)) {
530 struct gfs2_trans *tr = current->journal_info;
531 if (tr && tr->tr_touched)
532 gfs2_io_error_bh(sdp, bh);
535 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
543 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
544 * @gl: the glock the buffer belongs to
545 * @bh: The buffer to be attached to
546 * @meta: Flag to indicate whether its metadata or not
549 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
552 struct gfs2_bufdata *bd;
555 lock_page(bh->b_page);
559 unlock_page(bh->b_page);
563 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
564 memset(bd, 0, sizeof(struct gfs2_bufdata));
569 INIT_LIST_HEAD(&bd->bd_list_tr);
571 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
573 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
579 unlock_page(bh->b_page);
583 * gfs2_pin - Pin a buffer in memory
584 * @sdp: the filesystem the buffer belongs to
585 * @bh: The buffer to be pinned
589 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
591 struct gfs2_bufdata *bd = bh->b_private;
593 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
595 if (test_set_buffer_pinned(bh))
596 gfs2_assert_withdraw(sdp, 0);
600 /* If this buffer is in the AIL and it has already been written
601 to in-place disk block, remove it from the AIL. */
604 if (bd->bd_ail && !buffer_in_io(bh))
605 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
606 gfs2_log_unlock(sdp);
608 clear_buffer_dirty(bh);
611 if (!buffer_uptodate(bh))
612 gfs2_io_error_bh(sdp, bh);
618 * gfs2_unpin - Unpin a buffer
619 * @sdp: the filesystem the buffer belongs to
620 * @bh: The buffer to unpin
625 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
628 struct gfs2_bufdata *bd = bh->b_private;
630 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
632 if (!buffer_pinned(bh))
633 gfs2_assert_withdraw(sdp, 0);
635 mark_buffer_dirty(bh);
636 clear_buffer_pinned(bh);
640 list_del(&bd->bd_ail_st_list);
643 struct gfs2_glock *gl = bd->bd_gl;
644 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
645 atomic_inc(&gl->gl_ail_count);
648 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
649 gfs2_log_unlock(sdp);
653 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
654 * @ip: the inode who owns the buffers
655 * @bstart: the first buffer in the run
656 * @blen: the number of buffers in the run
660 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
662 struct gfs2_sbd *sdp = ip->i_sbd;
663 struct inode *aspace = ip->i_gl->gl_aspace;
664 struct buffer_head *bh;
667 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
669 struct gfs2_bufdata *bd = bh->b_private;
671 if (test_clear_buffer_pinned(bh)) {
672 struct gfs2_trans *tr = current->journal_info;
674 list_del_init(&bd->bd_le.le_list);
675 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
676 sdp->sd_log_num_buf--;
677 gfs2_log_unlock(sdp);
684 uint64_t blkno = bh->b_blocknr;
686 list_del(&bd->bd_ail_st_list);
687 list_del(&bd->bd_ail_gl_list);
688 atomic_dec(&bd->bd_gl->gl_ail_count);
690 gfs2_log_unlock(sdp);
691 gfs2_trans_add_revoke(sdp, blkno);
693 gfs2_log_unlock(sdp);
697 clear_buffer_dirty(bh);
698 clear_buffer_uptodate(bh);
710 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
711 * @ip: The GFS2 inode
713 * This releases buffers that are in the most-recently-used array of
714 * blocks used for indirect block addressing for this inode.
717 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
719 struct buffer_head **bh_slot;
722 spin_lock(&ip->i_spin);
724 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
725 bh_slot = &ip->i_cache[x];
732 spin_unlock(&ip->i_spin);
736 * gfs2_meta_indirect_buffer - Get a metadata buffer
737 * @ip: The GFS2 inode
738 * @height: The level of this buf in the metadata (indir addr) tree (if any)
739 * @num: The block number (device relative) of the buffer
740 * @new: Non-zero if we may create a new buffer
741 * @bhp: the buffer is returned here
743 * Try to use the gfs2_inode's MRU metadata tree cache.
748 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
749 int new, struct buffer_head **bhp)
751 struct buffer_head *bh, **bh_slot = ip->i_cache + height;
754 spin_lock(&ip->i_spin);
757 if (bh->b_blocknr == num)
762 spin_unlock(&ip->i_spin);
768 error = gfs2_meta_reread(ip->i_sbd, bh,
769 DIO_START | DIO_WAIT);
777 bh = gfs2_meta_new(ip->i_gl, num);
779 error = gfs2_meta_read(ip->i_gl, num,
780 DIO_START | DIO_WAIT, &bh);
785 spin_lock(&ip->i_spin);
786 if (*bh_slot != bh) {
791 spin_unlock(&ip->i_spin);
795 if (gfs2_assert_warn(ip->i_sbd, height)) {
799 gfs2_trans_add_bh(ip->i_gl, bh, 1);
800 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
801 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
803 } else if (gfs2_metatype_check(ip->i_sbd, bh,
804 (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
815 * gfs2_meta_ra - start readahead on an extent of a file
816 * @gl: the glock the blocks belong to
817 * @dblock: the starting disk block
818 * @extlen: the number of blocks in the extent
822 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
824 struct gfs2_sbd *sdp = gl->gl_sbd;
825 struct inode *aspace = gl->gl_aspace;
826 struct buffer_head *first_bh, *bh;
827 uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
828 sdp->sd_sb.sb_bsize_shift;
831 if (!extlen || !max_ra)
836 first_bh = getbuf(sdp, aspace, dblock, CREATE);
838 if (buffer_uptodate(first_bh))
840 if (!buffer_locked(first_bh)) {
841 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
850 bh = getbuf(sdp, aspace, dblock, CREATE);
852 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
853 error = gfs2_meta_reread(sdp, bh, DIO_START);
863 if (buffer_uptodate(first_bh))
872 * gfs2_meta_syncfs - sync all the buffers in a filesystem
873 * @sdp: the filesystem
877 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
879 gfs2_log_flush(sdp, NULL);
881 gfs2_ail1_start(sdp, DIO_ALL);
882 if (gfs2_ail1_empty(sdp, DIO_ALL))