blob: 718dbe5607ca744f902febbcf9bc0bbf6da9a5e0 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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/slab.h>
29#include <linux/highmem.h>
30
31#include <cluster/masklog.h>
32
33#include "ocfs2.h"
34
35#include "alloc.h"
36#include "inode.h"
37#include "journal.h"
38#include "uptodate.h"
39
40#include "buffer_head_io.h"
41
42int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
43 struct inode *inode)
44{
45 int ret = 0;
46
47 mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
48 (unsigned long long)bh->b_blocknr, inode);
49
50 BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
51 BUG_ON(buffer_jbd(bh));
52
53 /* No need to check for a soft readonly file system here. non
54 * journalled writes are only ever done on system files which
55 * can get modified during recovery even if read-only. */
56 if (ocfs2_is_hard_readonly(osb)) {
57 ret = -EROFS;
58 goto out;
59 }
60
Mark Fasheh251b6ec2006-01-10 15:41:43 -080061 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -080062
63 lock_buffer(bh);
64 set_buffer_uptodate(bh);
65
66 /* remove from dirty list before I/O. */
67 clear_buffer_dirty(bh);
68
Joel Beckerda1e9092008-10-09 17:20:29 -070069 get_bh(bh); /* for end_buffer_write_sync() */
Mark Fashehccd979b2005-12-15 14:31:24 -080070 bh->b_end_io = end_buffer_write_sync;
71 submit_bh(WRITE, bh);
72
73 wait_on_buffer(bh);
74
75 if (buffer_uptodate(bh)) {
76 ocfs2_set_buffer_uptodate(inode, bh);
77 } else {
78 /* We don't need to remove the clustered uptodate
79 * information for this bh as it's not marked locally
80 * uptodate. */
81 ret = -EIO;
Mark Fasheh2fe5c1d2008-01-23 18:35:31 -080082 put_bh(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -080083 }
84
Mark Fasheh251b6ec2006-01-10 15:41:43 -080085 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -080086out:
87 mlog_exit(ret);
88 return ret;
89}
90
Joel Beckerda1e9092008-10-09 17:20:29 -070091int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
92 unsigned int nr, struct buffer_head *bhs[])
93{
94 int status = 0;
95 unsigned int i;
96 struct buffer_head *bh;
97
98 if (!nr) {
99 mlog(ML_BH_IO, "No buffers will be read!\n");
100 goto bail;
101 }
102
103 for (i = 0 ; i < nr ; i++) {
104 if (bhs[i] == NULL) {
105 bhs[i] = sb_getblk(osb->sb, block++);
106 if (bhs[i] == NULL) {
107 status = -EIO;
108 mlog_errno(status);
109 goto bail;
110 }
111 }
112 bh = bhs[i];
113
114 if (buffer_jbd(bh)) {
115 mlog(ML_ERROR,
116 "trying to sync read a jbd "
117 "managed bh (blocknr = %llu), skipping\n",
118 (unsigned long long)bh->b_blocknr);
119 continue;
120 }
121
122 if (buffer_dirty(bh)) {
123 /* This should probably be a BUG, or
124 * at least return an error. */
125 mlog(ML_ERROR,
126 "trying to sync read a dirty "
127 "buffer! (blocknr = %llu), skipping\n",
128 (unsigned long long)bh->b_blocknr);
129 continue;
130 }
131
132 lock_buffer(bh);
133 if (buffer_jbd(bh)) {
134 mlog(ML_ERROR,
135 "block %llu had the JBD bit set "
136 "while I was in lock_buffer!",
137 (unsigned long long)bh->b_blocknr);
138 BUG();
139 }
140
141 clear_buffer_uptodate(bh);
142 get_bh(bh); /* for end_buffer_read_sync() */
143 bh->b_end_io = end_buffer_read_sync;
144 submit_bh(READ, bh);
145 }
146
147 for (i = nr; i > 0; i--) {
148 bh = bhs[i - 1];
149
150 if (buffer_jbd(bh)) {
151 mlog(ML_ERROR,
152 "the journal got the buffer while it was "
153 "locked for io! (blocknr = %llu)\n",
154 (unsigned long long)bh->b_blocknr);
155 BUG();
156 }
157
158 wait_on_buffer(bh);
159 if (!buffer_uptodate(bh)) {
160 /* Status won't be cleared from here on out,
161 * so we can safely record this and loop back
162 * to cleanup the other buffers. */
163 status = -EIO;
164 put_bh(bh);
165 bhs[i - 1] = NULL;
166 }
167 }
168
169bail:
170 return status;
171}
172
Joel Becker31d33072008-10-09 17:20:30 -0700173int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
174 struct buffer_head *bhs[], int flags)
Mark Fashehccd979b2005-12-15 14:31:24 -0800175{
176 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800177 int i, ignore_cache = 0;
178 struct buffer_head *bh;
179
Joel Becker31d33072008-10-09 17:20:30 -0700180 mlog_entry("(inode=%p, block=(%llu), nr=(%d), flags=%d)\n",
181 inode, (unsigned long long)block, nr, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800182
Joel Becker31d33072008-10-09 17:20:30 -0700183 BUG_ON(!inode);
184 BUG_ON((flags & OCFS2_BH_READAHEAD) && !(flags & OCFS2_BH_CACHED));
Mark Fashehaa958872006-04-21 13:49:02 -0700185
Joel Becker31d33072008-10-09 17:20:30 -0700186 if (bhs == NULL) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800187 status = -EINVAL;
188 mlog_errno(status);
189 goto bail;
190 }
191
192 if (nr < 0) {
193 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
194 status = -EINVAL;
195 mlog_errno(status);
196 goto bail;
197 }
198
199 if (nr == 0) {
200 mlog(ML_BH_IO, "No buffers will be read!\n");
201 status = 0;
202 goto bail;
203 }
204
Joel Becker31d33072008-10-09 17:20:30 -0700205 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800206 for (i = 0 ; i < nr ; i++) {
207 if (bhs[i] == NULL) {
Joel Becker31d33072008-10-09 17:20:30 -0700208 bhs[i] = sb_getblk(inode->i_sb, block++);
Mark Fashehccd979b2005-12-15 14:31:24 -0800209 if (bhs[i] == NULL) {
Joel Becker31d33072008-10-09 17:20:30 -0700210 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800211 status = -EIO;
212 mlog_errno(status);
213 goto bail;
214 }
215 }
216 bh = bhs[i];
217 ignore_cache = 0;
218
Mark Fashehaa958872006-04-21 13:49:02 -0700219 /* There are three read-ahead cases here which we need to
220 * be concerned with. All three assume a buffer has
221 * previously been submitted with OCFS2_BH_READAHEAD
222 * and it hasn't yet completed I/O.
223 *
224 * 1) The current request is sync to disk. This rarely
225 * happens these days, and never when performance
226 * matters - the code can just wait on the buffer
227 * lock and re-submit.
228 *
229 * 2) The current request is cached, but not
230 * readahead. ocfs2_buffer_uptodate() will return
231 * false anyway, so we'll wind up waiting on the
232 * buffer lock to do I/O. We re-check the request
233 * with after getting the lock to avoid a re-submit.
234 *
235 * 3) The current request is readahead (and so must
236 * also be a caching one). We short circuit if the
237 * buffer is locked (under I/O) and if it's in the
238 * uptodate cache. The re-check from #2 catches the
239 * case that the previous read-ahead completes just
240 * before our is-it-in-flight check.
241 */
242
Mark Fashehccd979b2005-12-15 14:31:24 -0800243 if (flags & OCFS2_BH_CACHED &&
244 !ocfs2_buffer_uptodate(inode, bh)) {
245 mlog(ML_UPTODATE,
Mark Fashehb06970532006-03-03 10:24:33 -0800246 "bh (%llu), inode %llu not uptodate\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800247 (unsigned long long)bh->b_blocknr,
Mark Fashehb06970532006-03-03 10:24:33 -0800248 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800249 ignore_cache = 1;
250 }
251
252 /* XXX: Can we ever get this and *not* have the cached
253 * flag set? */
254 if (buffer_jbd(bh)) {
255 if (!(flags & OCFS2_BH_CACHED) || ignore_cache)
256 mlog(ML_BH_IO, "trying to sync read a jbd "
257 "managed bh (blocknr = %llu)\n",
258 (unsigned long long)bh->b_blocknr);
259 continue;
260 }
261
262 if (!(flags & OCFS2_BH_CACHED) || ignore_cache) {
263 if (buffer_dirty(bh)) {
264 /* This should probably be a BUG, or
265 * at least return an error. */
266 mlog(ML_BH_IO, "asking me to sync read a dirty "
267 "buffer! (blocknr = %llu)\n",
268 (unsigned long long)bh->b_blocknr);
269 continue;
270 }
271
Mark Fashehaa958872006-04-21 13:49:02 -0700272 /* A read-ahead request was made - if the
273 * buffer is already under read-ahead from a
274 * previously submitted request than we are
275 * done here. */
276 if ((flags & OCFS2_BH_READAHEAD)
277 && ocfs2_buffer_read_ahead(inode, bh))
278 continue;
279
Mark Fashehccd979b2005-12-15 14:31:24 -0800280 lock_buffer(bh);
281 if (buffer_jbd(bh)) {
282#ifdef CATCH_BH_JBD_RACES
283 mlog(ML_ERROR, "block %llu had the JBD bit set "
284 "while I was in lock_buffer!",
285 (unsigned long long)bh->b_blocknr);
286 BUG();
287#else
288 unlock_buffer(bh);
289 continue;
290#endif
291 }
Mark Fashehaa958872006-04-21 13:49:02 -0700292
293 /* Re-check ocfs2_buffer_uptodate() as a
294 * previously read-ahead buffer may have
295 * completed I/O while we were waiting for the
296 * buffer lock. */
297 if ((flags & OCFS2_BH_CACHED)
298 && !(flags & OCFS2_BH_READAHEAD)
299 && ocfs2_buffer_uptodate(inode, bh)) {
300 unlock_buffer(bh);
301 continue;
302 }
303
Mark Fashehccd979b2005-12-15 14:31:24 -0800304 clear_buffer_uptodate(bh);
305 get_bh(bh); /* for end_buffer_read_sync() */
306 bh->b_end_io = end_buffer_read_sync;
Mark Fashehaa958872006-04-21 13:49:02 -0700307 submit_bh(READ, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800308 continue;
309 }
310 }
311
312 status = 0;
313
314 for (i = (nr - 1); i >= 0; i--) {
315 bh = bhs[i];
316
Mark Fashehaa958872006-04-21 13:49:02 -0700317 if (!(flags & OCFS2_BH_READAHEAD)) {
318 /* We know this can't have changed as we hold the
319 * inode sem. Avoid doing any work on the bh if the
320 * journal has it. */
321 if (!buffer_jbd(bh))
322 wait_on_buffer(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800323
Mark Fashehaa958872006-04-21 13:49:02 -0700324 if (!buffer_uptodate(bh)) {
325 /* Status won't be cleared from here on out,
326 * so we can safely record this and loop back
327 * to cleanup the other buffers. Don't need to
328 * remove the clustered uptodate information
329 * for this bh as it's not marked locally
330 * uptodate. */
331 status = -EIO;
Mark Fasheh2fe5c1d2008-01-23 18:35:31 -0800332 put_bh(bh);
Mark Fashehaa958872006-04-21 13:49:02 -0700333 bhs[i] = NULL;
334 continue;
335 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800336 }
337
Mark Fashehaa958872006-04-21 13:49:02 -0700338 /* Always set the buffer in the cache, even if it was
339 * a forced read, or read-ahead which hasn't yet
340 * completed. */
Joel Becker31d33072008-10-09 17:20:30 -0700341 ocfs2_set_buffer_uptodate(inode, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800342 }
Joel Becker31d33072008-10-09 17:20:30 -0700343 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800344
Mark Fashehaa958872006-04-21 13:49:02 -0700345 mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800346 (unsigned long long)block, nr,
Mark Fashehaa958872006-04-21 13:49:02 -0700347 (!(flags & OCFS2_BH_CACHED) || ignore_cache) ? "no" : "yes", flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800348
349bail:
350
351 mlog_exit(status);
352 return status;
353}
Tao Mad6590722007-12-18 15:47:03 +0800354
355/* Check whether the blkno is the super block or one of the backups. */
356static void ocfs2_check_super_or_backup(struct super_block *sb,
357 sector_t blkno)
358{
359 int i;
360 u64 backup_blkno;
361
362 if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
363 return;
364
365 for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
366 backup_blkno = ocfs2_backup_super_blkno(sb, i);
367 if (backup_blkno == blkno)
368 return;
369 }
370
371 BUG();
372}
373
374/*
375 * Write super block and backups doesn't need to collaborate with journal,
376 * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
377 * into this function.
378 */
379int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
380 struct buffer_head *bh)
381{
382 int ret = 0;
383
384 mlog_entry_void();
385
386 BUG_ON(buffer_jbd(bh));
387 ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
388
389 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
390 ret = -EROFS;
391 goto out;
392 }
393
394 lock_buffer(bh);
395 set_buffer_uptodate(bh);
396
397 /* remove from dirty list before I/O. */
398 clear_buffer_dirty(bh);
399
400 get_bh(bh); /* for end_buffer_write_sync() */
401 bh->b_end_io = end_buffer_write_sync;
402 submit_bh(WRITE, bh);
403
404 wait_on_buffer(bh);
405
406 if (!buffer_uptodate(bh)) {
407 ret = -EIO;
Mark Fasheh2fe5c1d2008-01-23 18:35:31 -0800408 put_bh(bh);
Tao Mad6590722007-12-18 15:47:03 +0800409 }
410
411out:
412 mlog_exit(ret);
413 return ret;
414}