[PATCH] OCFS2: The Second Oracle Cluster Filesystem
[linux-3.10.git] / fs / ocfs2 / buffer_head_io.c
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/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
42 int 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
61         down(&OCFS2_I(inode)->ip_io_sem);
62
63         lock_buffer(bh);
64         set_buffer_uptodate(bh);
65
66         /* remove from dirty list before I/O. */
67         clear_buffer_dirty(bh);
68
69         get_bh(bh); /* for end_buffer_write_sync() */                   
70         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;
82                 brelse(bh);
83         }
84
85         up(&OCFS2_I(inode)->ip_io_sem);
86 out:
87         mlog_exit(ret);
88         return ret;
89 }
90
91 int ocfs2_read_blocks(struct ocfs2_super *osb, u64 block, int nr,
92                       struct buffer_head *bhs[], int flags,
93                       struct inode *inode)
94 {
95         int status = 0;
96         struct super_block *sb;
97         int i, ignore_cache = 0;
98         struct buffer_head *bh;
99
100         mlog_entry("(block=(%"MLFu64"), nr=(%d), flags=%d, inode=%p)\n",
101                    block, nr, flags, inode);
102
103         if (osb == NULL || osb->sb == NULL || bhs == NULL) {
104                 status = -EINVAL;
105                 mlog_errno(status);
106                 goto bail;
107         }
108
109         if (nr < 0) {
110                 mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
111                 status = -EINVAL;
112                 mlog_errno(status);
113                 goto bail;
114         }
115
116         if (nr == 0) {
117                 mlog(ML_BH_IO, "No buffers will be read!\n");
118                 status = 0;
119                 goto bail;
120         }
121
122         sb = osb->sb;
123
124         if (flags & OCFS2_BH_CACHED && !inode)
125                 flags &= ~OCFS2_BH_CACHED;
126
127         if (inode)
128                 down(&OCFS2_I(inode)->ip_io_sem);
129         for (i = 0 ; i < nr ; i++) {
130                 if (bhs[i] == NULL) {
131                         bhs[i] = sb_getblk(sb, block++);
132                         if (bhs[i] == NULL) {
133                                 if (inode)
134                                         up(&OCFS2_I(inode)->ip_io_sem);
135                                 status = -EIO;
136                                 mlog_errno(status);
137                                 goto bail;
138                         }
139                 }
140                 bh = bhs[i];
141                 ignore_cache = 0;
142
143                 if (flags & OCFS2_BH_CACHED &&
144                     !ocfs2_buffer_uptodate(inode, bh)) {
145                         mlog(ML_UPTODATE,
146                              "bh (%llu), inode %"MLFu64" not uptodate\n",
147                              (unsigned long long)bh->b_blocknr,
148                              OCFS2_I(inode)->ip_blkno);
149                         ignore_cache = 1;
150                 }
151
152                 /* XXX: Can we ever get this and *not* have the cached
153                  * flag set? */
154                 if (buffer_jbd(bh)) {
155                         if (!(flags & OCFS2_BH_CACHED) || ignore_cache)
156                                 mlog(ML_BH_IO, "trying to sync read a jbd "
157                                                "managed bh (blocknr = %llu)\n",
158                                      (unsigned long long)bh->b_blocknr);
159                         continue;
160                 }
161
162                 if (!(flags & OCFS2_BH_CACHED) || ignore_cache) {
163                         if (buffer_dirty(bh)) {
164                                 /* This should probably be a BUG, or
165                                  * at least return an error. */
166                                 mlog(ML_BH_IO, "asking me to sync read a dirty "
167                                                "buffer! (blocknr = %llu)\n",
168                                      (unsigned long long)bh->b_blocknr);
169                                 continue;
170                         }
171
172                         lock_buffer(bh);
173                         if (buffer_jbd(bh)) {
174 #ifdef CATCH_BH_JBD_RACES
175                                 mlog(ML_ERROR, "block %llu had the JBD bit set "
176                                                "while I was in lock_buffer!",
177                                      (unsigned long long)bh->b_blocknr);
178                                 BUG();
179 #else
180                                 unlock_buffer(bh);
181                                 continue;
182 #endif
183                         }
184                         clear_buffer_uptodate(bh);
185                         get_bh(bh); /* for end_buffer_read_sync() */
186                         bh->b_end_io = end_buffer_read_sync;
187                         if (flags & OCFS2_BH_READAHEAD)
188                                 submit_bh(READA, bh);
189                         else
190                                 submit_bh(READ, bh);
191                         continue;
192                 }
193         }
194
195         status = 0;
196
197         for (i = (nr - 1); i >= 0; i--) {
198                 bh = bhs[i];
199
200                 /* We know this can't have changed as we hold the
201                  * inode sem. Avoid doing any work on the bh if the
202                  * journal has it. */
203                 if (!buffer_jbd(bh))
204                         wait_on_buffer(bh);
205
206                 if (!buffer_uptodate(bh)) {
207                         /* Status won't be cleared from here on out,
208                          * so we can safely record this and loop back
209                          * to cleanup the other buffers. Don't need to
210                          * remove the clustered uptodate information
211                          * for this bh as it's not marked locally
212                          * uptodate. */
213                         status = -EIO;
214                         brelse(bh);
215                         bhs[i] = NULL;
216                         continue;
217                 }
218
219                 if (inode)
220                         ocfs2_set_buffer_uptodate(inode, bh);
221         }
222         if (inode)
223                 up(&OCFS2_I(inode)->ip_io_sem);
224
225         mlog(ML_BH_IO, "block=(%"MLFu64"), nr=(%d), cached=%s\n", block, nr,
226              (!(flags & OCFS2_BH_CACHED) || ignore_cache) ? "no" : "yes");
227
228 bail:
229
230         mlog_exit(status);
231         return status;
232 }