1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * io.c
5 *
6 * Buffer cache handling
7 *
8 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/highmem.h>
29
30#include <cluster/masklog.h>
31
32#include "ocfs2.h"
33
34#include "alloc.h"
35#include "inode.h"
36#include "journal.h"
37#include "uptodate.h"
38#include "buffer_head_io.h"
39#include "ocfs2_trace.h"
40
41/*
42 * Bits on bh->b_state used by ocfs2.
43 *
44 * These MUST be after the JBD2 bits.  Hence, we use BH_JBDPrivateStart.
45 */
46enum ocfs2_state_bits {
47	BH_NeedsValidate = BH_JBDPrivateStart,
48};
49
50/* Expand the magic b_state functions */
51BUFFER_FNS(NeedsValidate, needs_validate);
52
53int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
54		      struct ocfs2_caching_info *ci)
55{
56	int ret = 0;
57
58	trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
59
60	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
61	BUG_ON(buffer_jbd(bh));
62
63	/* No need to check for a soft readonly file system here. non
64	 * journalled writes are only ever done on system files which
65	 * can get modified during recovery even if read-only. */
66	if (ocfs2_is_hard_readonly(osb)) {
67		ret = -EROFS;
68		mlog_errno(ret);
69		goto out;
70	}
71
72	ocfs2_metadata_cache_io_lock(ci);
73
74	lock_buffer(bh);
75	set_buffer_uptodate(bh);
76
77	/* remove from dirty list before I/O. */
78	clear_buffer_dirty(bh);
79
80	get_bh(bh); /* for end_buffer_write_sync() */
81	bh->b_end_io = end_buffer_write_sync;
82	submit_bh(WRITE, bh);
83
84	wait_on_buffer(bh);
85
86	if (buffer_uptodate(bh)) {
87		ocfs2_set_buffer_uptodate(ci, bh);
88	} else {
89		/* We don't need to remove the clustered uptodate
90		 * information for this bh as it's not marked locally
91		 * uptodate. */
92		ret = -EIO;
93		mlog_errno(ret);
94	}
95
96	ocfs2_metadata_cache_io_unlock(ci);
97out:
98	return ret;
99}
100
101int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
102			   unsigned int nr, struct buffer_head *bhs[])
103{
104	int status = 0;
105	unsigned int i;
106	struct buffer_head *bh;
107
108	trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
109
110	if (!nr)
111		goto bail;
112
113	for (i = 0 ; i < nr ; i++) {
114		if (bhs[i] == NULL) {
115			bhs[i] = sb_getblk(osb->sb, block++);
116			if (bhs[i] == NULL) {
117				status = -ENOMEM;
118				mlog_errno(status);
119				goto bail;
120			}
121		}
122		bh = bhs[i];
123
124		if (buffer_jbd(bh)) {
125			trace_ocfs2_read_blocks_sync_jbd(
126					(unsigned long long)bh->b_blocknr);
127			continue;
128		}
129
130		if (buffer_dirty(bh)) {
131			/* This should probably be a BUG, or
132			 * at least return an error. */
133			mlog(ML_ERROR,
134			     "trying to sync read a dirty "
135			     "buffer! (blocknr = %llu), skipping\n",
136			     (unsigned long long)bh->b_blocknr);
137			continue;
138		}
139
140		lock_buffer(bh);
141		if (buffer_jbd(bh)) {
142			mlog(ML_ERROR,
143			     "block %llu had the JBD bit set "
144			     "while I was in lock_buffer!",
145			     (unsigned long long)bh->b_blocknr);
146			BUG();
147		}
148
149		clear_buffer_uptodate(bh);
150		get_bh(bh); /* for end_buffer_read_sync() */
151		bh->b_end_io = end_buffer_read_sync;
152		submit_bh(READ, bh);
153	}
154
155	for (i = nr; i > 0; i--) {
156		bh = bhs[i - 1];
157
158		/* No need to wait on the buffer if it's managed by JBD. */
159		if (!buffer_jbd(bh))
160			wait_on_buffer(bh);
161
162		if (!buffer_uptodate(bh)) {
163			/* Status won't be cleared from here on out,
164			 * so we can safely record this and loop back
165			 * to cleanup the other buffers. */
166			status = -EIO;
167			put_bh(bh);
168			bhs[i - 1] = NULL;
169		}
170	}
171
172bail:
173	return status;
174}
175
176int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
177		      struct buffer_head *bhs[], int flags,
178		      int (*validate)(struct super_block *sb,
179				      struct buffer_head *bh))
180{
181	int status = 0;
182	int i, ignore_cache = 0;
183	struct buffer_head *bh;
184	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
185
186	trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
187
188	BUG_ON(!ci);
189	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
190	       (flags & OCFS2_BH_IGNORE_CACHE));
191
192	if (bhs == NULL) {
193		status = -EINVAL;
194		mlog_errno(status);
195		goto bail;
196	}
197
198	if (nr < 0) {
199		mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
200		status = -EINVAL;
201		mlog_errno(status);
202		goto bail;
203	}
204
205	if (nr == 0) {
206		status = 0;
207		goto bail;
208	}
209
210	ocfs2_metadata_cache_io_lock(ci);
211	for (i = 0 ; i < nr ; i++) {
212		if (bhs[i] == NULL) {
213			bhs[i] = sb_getblk(sb, block++);
214			if (bhs[i] == NULL) {
215				ocfs2_metadata_cache_io_unlock(ci);
216				status = -ENOMEM;
217				mlog_errno(status);
218				goto bail;
219			}
220		}
221		bh = bhs[i];
222		ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
223
224		/* There are three read-ahead cases here which we need to
225		 * be concerned with. All three assume a buffer has
226		 * previously been submitted with OCFS2_BH_READAHEAD
227		 * and it hasn't yet completed I/O.
228		 *
229		 * 1) The current request is sync to disk. This rarely
230		 *    happens these days, and never when performance
231		 *    matters - the code can just wait on the buffer
232		 *    lock and re-submit.
233		 *
234		 * 2) The current request is cached, but not
235		 *    readahead. ocfs2_buffer_uptodate() will return
236		 *    false anyway, so we'll wind up waiting on the
237		 *    buffer lock to do I/O. We re-check the request
238		 *    with after getting the lock to avoid a re-submit.
239		 *
240		 * 3) The current request is readahead (and so must
241		 *    also be a caching one). We short circuit if the
242		 *    buffer is locked (under I/O) and if it's in the
243		 *    uptodate cache. The re-check from #2 catches the
244		 *    case that the previous read-ahead completes just
245		 *    before our is-it-in-flight check.
246		 */
247
248		if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
249			trace_ocfs2_read_blocks_from_disk(
250			     (unsigned long long)bh->b_blocknr,
251			     (unsigned long long)ocfs2_metadata_cache_owner(ci));
252			/* We're using ignore_cache here to say
253			 * "go to disk" */
254			ignore_cache = 1;
255		}
256
257		trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
258			ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
259
260		if (buffer_jbd(bh)) {
261			continue;
262		}
263
264		if (ignore_cache) {
265			if (buffer_dirty(bh)) {
266				/* This should probably be a BUG, or
267				 * at least return an error. */
268				continue;
269			}
270
271			/* A read-ahead request was made - if the
272			 * buffer is already under read-ahead from a
273			 * previously submitted request than we are
274			 * done here. */
275			if ((flags & OCFS2_BH_READAHEAD)
276			    && ocfs2_buffer_read_ahead(ci, bh))
277				continue;
278
279			lock_buffer(bh);
280			if (buffer_jbd(bh)) {
281#ifdef CATCH_BH_JBD_RACES
282				mlog(ML_ERROR, "block %llu had the JBD bit set "
283					       "while I was in lock_buffer!",
284				     (unsigned long long)bh->b_blocknr);
285				BUG();
286#else
287				unlock_buffer(bh);
288				continue;
289#endif
290			}
291
292			/* Re-check ocfs2_buffer_uptodate() as a
293			 * previously read-ahead buffer may have
294			 * completed I/O while we were waiting for the
295			 * buffer lock. */
296			if (!(flags & OCFS2_BH_IGNORE_CACHE)
297			    && !(flags & OCFS2_BH_READAHEAD)
298			    && ocfs2_buffer_uptodate(ci, bh)) {
299				unlock_buffer(bh);
300				continue;
301			}
302
303			clear_buffer_uptodate(bh);
304			get_bh(bh); /* for end_buffer_read_sync() */
305			if (validate)
306				set_buffer_needs_validate(bh);
307			bh->b_end_io = end_buffer_read_sync;
308			submit_bh(READ, bh);
309			continue;
310		}
311	}
312
313	status = 0;
314
315	for (i = (nr - 1); i >= 0; i--) {
316		bh = bhs[i];
317
318		if (!(flags & OCFS2_BH_READAHEAD)) {
319			/* We know this can't have changed as we hold the
320			 * owner sem. Avoid doing any work on the bh if the
321			 * journal has it. */
322			if (!buffer_jbd(bh))
323				wait_on_buffer(bh);
324
325			if (!buffer_uptodate(bh)) {
326				/* Status won't be cleared from here on out,
327				 * so we can safely record this and loop back
328				 * to cleanup the other buffers. Don't need to
329				 * remove the clustered uptodate information
330				 * for this bh as it's not marked locally
331				 * uptodate. */
332				status = -EIO;
333				put_bh(bh);
334				bhs[i] = NULL;
335				continue;
336			}
337
338			if (buffer_needs_validate(bh)) {
339				/* We never set NeedsValidate if the
340				 * buffer was held by the journal, so
341				 * that better not have changed */
342				BUG_ON(buffer_jbd(bh));
343				clear_buffer_needs_validate(bh);
344				status = validate(sb, bh);
345				if (status) {
346					put_bh(bh);
347					bhs[i] = NULL;
348					continue;
349				}
350			}
351		}
352
353		/* Always set the buffer in the cache, even if it was
354		 * a forced read, or read-ahead which hasn't yet
355		 * completed. */
356		ocfs2_set_buffer_uptodate(ci, bh);
357	}
358	ocfs2_metadata_cache_io_unlock(ci);
359
360	trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
361				    flags, ignore_cache);
362
363bail:
364
365	return status;
366}
367
368/* Check whether the blkno is the super block or one of the backups. */
369static void ocfs2_check_super_or_backup(struct super_block *sb,
370					sector_t blkno)
371{
372	int i;
373	u64 backup_blkno;
374
375	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
376		return;
377
378	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
379		backup_blkno = ocfs2_backup_super_blkno(sb, i);
380		if (backup_blkno == blkno)
381			return;
382	}
383
384	BUG();
385}
386
387/*
388 * Write super block and backups doesn't need to collaborate with journal,
389 * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
390 * into this function.
391 */
392int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
393				struct buffer_head *bh)
394{
395	int ret = 0;
396	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
397
398	BUG_ON(buffer_jbd(bh));
399	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
400
401	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
402		ret = -EROFS;
403		mlog_errno(ret);
404		goto out;
405	}
406
407	lock_buffer(bh);
408	set_buffer_uptodate(bh);
409
410	/* remove from dirty list before I/O. */
411	clear_buffer_dirty(bh);
412
413	get_bh(bh); /* for end_buffer_write_sync() */
414	bh->b_end_io = end_buffer_write_sync;
415	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
416	submit_bh(WRITE, bh);
417
418	wait_on_buffer(bh);
419
420	if (!buffer_uptodate(bh)) {
421		ret = -EIO;
422		mlog_errno(ret);
423	}
424
425out:
426	return ret;
427}
428