2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 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>
34 #include "trace_gfs2.h"
36 static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
38 struct buffer_head *bh, *head;
40 int write_op = REQ_META | REQ_PRIO |
41 (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
43 BUG_ON(!PageLocked(page));
44 BUG_ON(!page_has_buffers(page));
46 head = page_buffers(page);
50 if (!buffer_mapped(bh))
53 * If it's a fully non-blocking write attempt and we cannot
54 * lock the buffer then redirty the page. Note that this can
55 * potentially cause a busy-wait loop from flusher thread and kswapd
56 * activity, but those code paths have their own higher-level
59 if (wbc->sync_mode != WB_SYNC_NONE) {
61 } else if (!trylock_buffer(bh)) {
62 redirty_page_for_writepage(wbc, page);
65 if (test_clear_buffer_dirty(bh)) {
66 mark_buffer_async_write(bh);
70 } while ((bh = bh->b_this_page) != head);
73 * The page and its buffers are protected by PageWriteback(), so we can
74 * drop the bh refcounts early.
76 BUG_ON(PageWriteback(page));
77 set_page_writeback(page);
80 struct buffer_head *next = bh->b_this_page;
81 if (buffer_async_write(bh)) {
82 submit_bh(write_op, bh);
90 end_page_writeback(page);
95 const struct address_space_operations gfs2_meta_aops = {
96 .writepage = gfs2_aspace_writepage,
97 .releasepage = gfs2_releasepage,
101 * gfs2_meta_sync - Sync all buffers associated with a glock
106 void gfs2_meta_sync(struct gfs2_glock *gl)
108 struct address_space *mapping = gfs2_glock2aspace(gl);
111 filemap_fdatawrite(mapping);
112 error = filemap_fdatawait(mapping);
115 gfs2_io_error(gl->gl_sbd);
119 * gfs2_getbuf - Get a buffer with a given address space
121 * @blkno: the block number (filesystem scope)
122 * @create: 1 if the buffer should be created
124 * Returns: the buffer
127 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
129 struct address_space *mapping = gfs2_glock2aspace(gl);
130 struct gfs2_sbd *sdp = gl->gl_sbd;
132 struct buffer_head *bh;
137 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
138 index = blkno >> shift; /* convert block to page */
139 bufnum = blkno - (index << shift); /* block buf index within page */
143 page = grab_cache_page(mapping, index);
149 page = find_lock_page(mapping, index);
154 if (!page_has_buffers(page))
155 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
157 /* Locate header for our buffer within our page */
158 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
162 if (!buffer_mapped(bh))
163 map_bh(bh, sdp->sd_vfs, blkno);
166 mark_page_accessed(page);
167 page_cache_release(page);
172 static void meta_prep_new(struct buffer_head *bh)
174 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
177 clear_buffer_dirty(bh);
178 set_buffer_uptodate(bh);
181 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
185 * gfs2_meta_new - Get a block
186 * @gl: The glock associated with this block
187 * @blkno: The block number
189 * Returns: The buffer
192 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
194 struct buffer_head *bh;
195 bh = gfs2_getbuf(gl, blkno, CREATE);
201 * gfs2_meta_read - Read a block from disk
202 * @gl: The glock covering the block
203 * @blkno: The block number
205 * @bhp: the place where the buffer is returned (NULL on failure)
210 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
211 struct buffer_head **bhp)
213 struct gfs2_sbd *sdp = gl->gl_sbd;
214 struct buffer_head *bh;
216 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
221 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
224 if (buffer_uptodate(bh)) {
228 bh->b_end_io = end_buffer_read_sync;
230 submit_bh(READ_SYNC | REQ_META | REQ_PRIO, bh);
231 if (!(flags & DIO_WAIT))
235 if (unlikely(!buffer_uptodate(bh))) {
236 struct gfs2_trans *tr = current->journal_info;
237 if (tr && tr->tr_touched)
238 gfs2_io_error_bh(sdp, bh);
248 * gfs2_meta_wait - Reread a block from disk
249 * @sdp: the filesystem
250 * @bh: The block to wait for
255 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
257 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
262 if (!buffer_uptodate(bh)) {
263 struct gfs2_trans *tr = current->journal_info;
264 if (tr && tr->tr_touched)
265 gfs2_io_error_bh(sdp, bh);
268 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
274 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
276 struct address_space *mapping = bh->b_page->mapping;
277 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
278 struct gfs2_bufdata *bd = bh->b_private;
280 if (test_clear_buffer_pinned(bh)) {
281 trace_gfs2_pin(bd, 0);
282 atomic_dec(&sdp->sd_log_pinned);
283 list_del_init(&bd->bd_list);
285 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
286 sdp->sd_log_num_buf--;
289 gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
290 sdp->sd_log_num_databuf--;
291 tr->tr_num_databuf_rm++;
297 spin_lock(&sdp->sd_ail_lock);
299 gfs2_remove_from_ail(bd);
300 bh->b_private = NULL;
302 bd->bd_blkno = bh->b_blocknr;
303 gfs2_trans_add_revoke(sdp, bd);
305 spin_unlock(&sdp->sd_ail_lock);
307 clear_buffer_dirty(bh);
308 clear_buffer_uptodate(bh);
312 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
313 * @ip: the inode who owns the buffers
314 * @bstart: the first buffer in the run
315 * @blen: the number of buffers in the run
319 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
321 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
322 struct buffer_head *bh;
325 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
329 gfs2_remove_from_journal(bh, current->journal_info, 1);
330 gfs2_log_unlock(sdp);
341 * gfs2_meta_indirect_buffer - Get a metadata buffer
342 * @ip: The GFS2 inode
343 * @height: The level of this buf in the metadata (indir addr) tree (if any)
344 * @num: The block number (device relative) of the buffer
345 * @bhp: the buffer is returned here
350 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
351 struct buffer_head **bhp)
353 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
354 struct gfs2_glock *gl = ip->i_gl;
355 struct buffer_head *bh;
357 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
359 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
360 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
369 * gfs2_meta_ra - start readahead on an extent of a file
370 * @gl: the glock the blocks belong to
371 * @dblock: the starting disk block
372 * @extlen: the number of blocks in the extent
374 * returns: the first buffer in the extent
377 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
379 struct gfs2_sbd *sdp = gl->gl_sbd;
380 struct buffer_head *first_bh, *bh;
381 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
382 sdp->sd_sb.sb_bsize_shift;
391 first_bh = gfs2_getbuf(gl, dblock, CREATE);
393 if (buffer_uptodate(first_bh))
395 if (!buffer_locked(first_bh))
396 ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh);
402 bh = gfs2_getbuf(gl, dblock, CREATE);
404 if (!buffer_uptodate(bh) && !buffer_locked(bh))
405 ll_rw_block(READA | REQ_META, 1, &bh);
409 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
413 wait_on_buffer(first_bh);