1/*
2 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4 *
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.
8 */
9
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>
15#include <linux/mm.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
23#include "gfs2.h"
24#include "incore.h"
25#include "glock.h"
26#include "glops.h"
27#include "inode.h"
28#include "log.h"
29#include "lops.h"
30#include "meta_io.h"
31#include "rgrp.h"
32#include "trans.h"
33#include "util.h"
34#include "trace_gfs2.h"
35
36static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
37{
38	struct buffer_head *bh, *head;
39	int nr_underway = 0;
40	int write_op = REQ_META | REQ_PRIO |
41		(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
42
43	BUG_ON(!PageLocked(page));
44	BUG_ON(!page_has_buffers(page));
45
46	head = page_buffers(page);
47	bh = head;
48
49	do {
50		if (!buffer_mapped(bh))
51			continue;
52		/*
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
57		 * throttling.
58		 */
59		if (wbc->sync_mode != WB_SYNC_NONE) {
60			lock_buffer(bh);
61		} else if (!trylock_buffer(bh)) {
62			redirty_page_for_writepage(wbc, page);
63			continue;
64		}
65		if (test_clear_buffer_dirty(bh)) {
66			mark_buffer_async_write(bh);
67		} else {
68			unlock_buffer(bh);
69		}
70	} while ((bh = bh->b_this_page) != head);
71
72	/*
73	 * The page and its buffers are protected by PageWriteback(), so we can
74	 * drop the bh refcounts early.
75	 */
76	BUG_ON(PageWriteback(page));
77	set_page_writeback(page);
78
79	do {
80		struct buffer_head *next = bh->b_this_page;
81		if (buffer_async_write(bh)) {
82			submit_bh(write_op, bh);
83			nr_underway++;
84		}
85		bh = next;
86	} while (bh != head);
87	unlock_page(page);
88
89	if (nr_underway == 0)
90		end_page_writeback(page);
91
92	return 0;
93}
94
95const struct address_space_operations gfs2_meta_aops = {
96	.writepage = gfs2_aspace_writepage,
97	.releasepage = gfs2_releasepage,
98};
99
100const struct address_space_operations gfs2_rgrp_aops = {
101	.writepage = gfs2_aspace_writepage,
102	.releasepage = gfs2_releasepage,
103};
104
105/**
106 * gfs2_getbuf - Get a buffer with a given address space
107 * @gl: the glock
108 * @blkno: the block number (filesystem scope)
109 * @create: 1 if the buffer should be created
110 *
111 * Returns: the buffer
112 */
113
114struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
115{
116	struct address_space *mapping = gfs2_glock2aspace(gl);
117	struct gfs2_sbd *sdp = gl->gl_sbd;
118	struct page *page;
119	struct buffer_head *bh;
120	unsigned int shift;
121	unsigned long index;
122	unsigned int bufnum;
123
124	if (mapping == NULL)
125		mapping = &sdp->sd_aspace;
126
127	shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
128	index = blkno >> shift;             /* convert block to page */
129	bufnum = blkno - (index << shift);  /* block buf index within page */
130
131	if (create) {
132		for (;;) {
133			page = grab_cache_page(mapping, index);
134			if (page)
135				break;
136			yield();
137		}
138	} else {
139		page = find_get_page_flags(mapping, index,
140						FGP_LOCK|FGP_ACCESSED);
141		if (!page)
142			return NULL;
143	}
144
145	if (!page_has_buffers(page))
146		create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
147
148	/* Locate header for our buffer within our page */
149	for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
150		/* Do nothing */;
151	get_bh(bh);
152
153	if (!buffer_mapped(bh))
154		map_bh(bh, sdp->sd_vfs, blkno);
155
156	unlock_page(page);
157	page_cache_release(page);
158
159	return bh;
160}
161
162static void meta_prep_new(struct buffer_head *bh)
163{
164	struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
165
166	lock_buffer(bh);
167	clear_buffer_dirty(bh);
168	set_buffer_uptodate(bh);
169	unlock_buffer(bh);
170
171	mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
172}
173
174/**
175 * gfs2_meta_new - Get a block
176 * @gl: The glock associated with this block
177 * @blkno: The block number
178 *
179 * Returns: The buffer
180 */
181
182struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
183{
184	struct buffer_head *bh;
185	bh = gfs2_getbuf(gl, blkno, CREATE);
186	meta_prep_new(bh);
187	return bh;
188}
189
190/**
191 * gfs2_meta_read - Read a block from disk
192 * @gl: The glock covering the block
193 * @blkno: The block number
194 * @flags: flags
195 * @bhp: the place where the buffer is returned (NULL on failure)
196 *
197 * Returns: errno
198 */
199
200int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
201		   struct buffer_head **bhp)
202{
203	struct gfs2_sbd *sdp = gl->gl_sbd;
204	struct buffer_head *bh;
205
206	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
207		*bhp = NULL;
208		return -EIO;
209	}
210
211	*bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
212
213	lock_buffer(bh);
214	if (buffer_uptodate(bh)) {
215		unlock_buffer(bh);
216		return 0;
217	}
218	bh->b_end_io = end_buffer_read_sync;
219	get_bh(bh);
220	submit_bh(READ_SYNC | REQ_META | REQ_PRIO, bh);
221	if (!(flags & DIO_WAIT))
222		return 0;
223
224	wait_on_buffer(bh);
225	if (unlikely(!buffer_uptodate(bh))) {
226		struct gfs2_trans *tr = current->journal_info;
227		if (tr && tr->tr_touched)
228			gfs2_io_error_bh(sdp, bh);
229		brelse(bh);
230		*bhp = NULL;
231		return -EIO;
232	}
233
234	return 0;
235}
236
237/**
238 * gfs2_meta_wait - Reread a block from disk
239 * @sdp: the filesystem
240 * @bh: The block to wait for
241 *
242 * Returns: errno
243 */
244
245int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
246{
247	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
248		return -EIO;
249
250	wait_on_buffer(bh);
251
252	if (!buffer_uptodate(bh)) {
253		struct gfs2_trans *tr = current->journal_info;
254		if (tr && tr->tr_touched)
255			gfs2_io_error_bh(sdp, bh);
256		return -EIO;
257	}
258	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
259		return -EIO;
260
261	return 0;
262}
263
264void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
265{
266	struct address_space *mapping = bh->b_page->mapping;
267	struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
268	struct gfs2_bufdata *bd = bh->b_private;
269	int was_pinned = 0;
270
271	if (test_clear_buffer_pinned(bh)) {
272		trace_gfs2_pin(bd, 0);
273		atomic_dec(&sdp->sd_log_pinned);
274		list_del_init(&bd->bd_list);
275		if (meta)
276			tr->tr_num_buf_rm++;
277		else
278			tr->tr_num_databuf_rm++;
279		tr->tr_touched = 1;
280		was_pinned = 1;
281		brelse(bh);
282	}
283	if (bd) {
284		spin_lock(&sdp->sd_ail_lock);
285		if (bd->bd_tr) {
286			gfs2_trans_add_revoke(sdp, bd);
287		} else if (was_pinned) {
288			bh->b_private = NULL;
289			kmem_cache_free(gfs2_bufdata_cachep, bd);
290		}
291		spin_unlock(&sdp->sd_ail_lock);
292	}
293	clear_buffer_dirty(bh);
294	clear_buffer_uptodate(bh);
295}
296
297/**
298 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
299 * @ip: the inode who owns the buffers
300 * @bstart: the first buffer in the run
301 * @blen: the number of buffers in the run
302 *
303 */
304
305void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
306{
307	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
308	struct buffer_head *bh;
309
310	while (blen) {
311		bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
312		if (bh) {
313			lock_buffer(bh);
314			gfs2_log_lock(sdp);
315			gfs2_remove_from_journal(bh, current->journal_info, 1);
316			gfs2_log_unlock(sdp);
317			unlock_buffer(bh);
318			brelse(bh);
319		}
320
321		bstart++;
322		blen--;
323	}
324}
325
326/**
327 * gfs2_meta_indirect_buffer - Get a metadata buffer
328 * @ip: The GFS2 inode
329 * @height: The level of this buf in the metadata (indir addr) tree (if any)
330 * @num: The block number (device relative) of the buffer
331 * @bhp: the buffer is returned here
332 *
333 * Returns: errno
334 */
335
336int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
337			      struct buffer_head **bhp)
338{
339	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
340	struct gfs2_glock *gl = ip->i_gl;
341	struct buffer_head *bh;
342	int ret = 0;
343	u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
344
345	ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
346	if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
347		brelse(bh);
348		ret = -EIO;
349	}
350	*bhp = bh;
351	return ret;
352}
353
354/**
355 * gfs2_meta_ra - start readahead on an extent of a file
356 * @gl: the glock the blocks belong to
357 * @dblock: the starting disk block
358 * @extlen: the number of blocks in the extent
359 *
360 * returns: the first buffer in the extent
361 */
362
363struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
364{
365	struct gfs2_sbd *sdp = gl->gl_sbd;
366	struct buffer_head *first_bh, *bh;
367	u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
368			  sdp->sd_sb.sb_bsize_shift;
369
370	BUG_ON(!extlen);
371
372	if (max_ra < 1)
373		max_ra = 1;
374	if (extlen > max_ra)
375		extlen = max_ra;
376
377	first_bh = gfs2_getbuf(gl, dblock, CREATE);
378
379	if (buffer_uptodate(first_bh))
380		goto out;
381	if (!buffer_locked(first_bh))
382		ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh);
383
384	dblock++;
385	extlen--;
386
387	while (extlen) {
388		bh = gfs2_getbuf(gl, dblock, CREATE);
389
390		if (!buffer_uptodate(bh) && !buffer_locked(bh))
391			ll_rw_block(READA | REQ_META, 1, &bh);
392		brelse(bh);
393		dblock++;
394		extlen--;
395		if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
396			goto out;
397	}
398
399	wait_on_buffer(first_bh);
400out:
401	return first_bh;
402}
403
404