Merge branch 'for-linus' from kernel.org:/.../shaggy/jfs-2.6 manually
[pandora-kernel.git] / fs / jfs / inode.c
1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2004
3  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
4  *
5  *   This program is free software;  you can redistribute it and/or modify
6  *   it under the terms of the GNU General Public License as published by
7  *   the Free Software Foundation; either version 2 of the License, or 
8  *   (at your option) any later version.
9  * 
10  *   This program is distributed in the hope that it will be useful,
11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
13  *   the GNU General Public License for more details.
14  *
15  *   You should have received a copy of the GNU General Public License
16  *   along with this program;  if not, write to the Free Software 
17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18  */
19
20 #include <linux/fs.h>
21 #include <linux/mpage.h>
22 #include <linux/buffer_head.h>
23 #include <linux/pagemap.h>
24 #include <linux/quotaops.h>
25 #include "jfs_incore.h"
26 #include "jfs_inode.h"
27 #include "jfs_filsys.h"
28 #include "jfs_imap.h"
29 #include "jfs_extent.h"
30 #include "jfs_unicode.h"
31 #include "jfs_debug.h"
32
33
34 void jfs_read_inode(struct inode *inode)
35 {
36         if (diRead(inode)) { 
37                 make_bad_inode(inode);
38                 return;
39         }
40
41         if (S_ISREG(inode->i_mode)) {
42                 inode->i_op = &jfs_file_inode_operations;
43                 inode->i_fop = &jfs_file_operations;
44                 inode->i_mapping->a_ops = &jfs_aops;
45         } else if (S_ISDIR(inode->i_mode)) {
46                 inode->i_op = &jfs_dir_inode_operations;
47                 inode->i_fop = &jfs_dir_operations;
48         } else if (S_ISLNK(inode->i_mode)) {
49                 if (inode->i_size >= IDATASIZE) {
50                         inode->i_op = &page_symlink_inode_operations;
51                         inode->i_mapping->a_ops = &jfs_aops;
52                 } else
53                         inode->i_op = &jfs_symlink_inode_operations;
54         } else {
55                 inode->i_op = &jfs_file_inode_operations;
56                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
57         }
58 }
59
60 /*
61  * Workhorse of both fsync & write_inode
62  */
63 int jfs_commit_inode(struct inode *inode, int wait)
64 {
65         int rc = 0;
66         tid_t tid;
67         static int noisy = 5;
68
69         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
70
71         /*
72          * Don't commit if inode has been committed since last being
73          * marked dirty, or if it has been deleted.
74          */
75         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
76                 return 0;
77
78         if (isReadOnly(inode)) {
79                 /* kernel allows writes to devices on read-only
80                  * partitions and may think inode is dirty
81                  */
82                 if (!special_file(inode->i_mode) && noisy) {
83                         jfs_err("jfs_commit_inode(0x%p) called on "
84                                    "read-only volume", inode);
85                         jfs_err("Is remount racy?");
86                         noisy--;
87                 }
88                 return 0;
89         }
90
91         tid = txBegin(inode->i_sb, COMMIT_INODE);
92         down(&JFS_IP(inode)->commit_sem);
93
94         /*
95          * Retest inode state after taking commit_sem
96          */
97         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
98                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
99
100         txEnd(tid);
101         up(&JFS_IP(inode)->commit_sem);
102         return rc;
103 }
104
105 int jfs_write_inode(struct inode *inode, int wait)
106 {
107         if (test_cflag(COMMIT_Nolink, inode))
108                 return 0;
109         /*
110          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
111          * It has been committed since the last change, but was still
112          * on the dirty inode list.
113          */
114          if (!test_cflag(COMMIT_Dirty, inode)) {
115                 /* Make sure committed changes hit the disk */
116                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
117                 return 0;
118          }
119
120         if (jfs_commit_inode(inode, wait)) {
121                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
122                 return -EIO;
123         } else
124                 return 0;
125 }
126
127 void jfs_delete_inode(struct inode *inode)
128 {
129         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
130
131         if (!is_bad_inode(inode) &&
132             (JFS_IP(inode)->fileset == cpu_to_le32(FILESYSTEM_I))) {
133
134                 truncate_inode_pages(&inode->i_data, 0);
135
136                 if (test_cflag(COMMIT_Freewmap, inode))
137                         jfs_free_zero_link(inode);
138
139                 diFree(inode);
140
141                 /*
142                  * Free the inode from the quota allocation.
143                  */
144                 DQUOT_INIT(inode);
145                 DQUOT_FREE_INODE(inode);
146                 DQUOT_DROP(inode);
147         }
148
149         clear_inode(inode);
150 }
151
152 void jfs_dirty_inode(struct inode *inode)
153 {
154         static int noisy = 5;
155
156         if (isReadOnly(inode)) {
157                 if (!special_file(inode->i_mode) && noisy) {
158                         /* kernel allows writes to devices on read-only
159                          * partitions and may try to mark inode dirty
160                          */
161                         jfs_err("jfs_dirty_inode called on read-only volume");
162                         jfs_err("Is remount racy?");
163                         noisy--;
164                 }
165                 return;
166         }
167
168         set_cflag(COMMIT_Dirty, inode);
169 }
170
171 static int
172 jfs_get_blocks(struct inode *ip, sector_t lblock, unsigned long max_blocks,
173                         struct buffer_head *bh_result, int create)
174 {
175         s64 lblock64 = lblock;
176         int rc = 0;
177         xad_t xad;
178         s64 xaddr;
179         int xflag;
180         s32 xlen = max_blocks;
181
182         /*
183          * Take appropriate lock on inode
184          */
185         if (create)
186                 IWRITE_LOCK(ip);
187         else
188                 IREAD_LOCK(ip);
189
190         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
191             (!xtLookup(ip, lblock64, max_blocks, &xflag, &xaddr, &xlen, 0)) &&
192             xaddr) {
193                 if (xflag & XAD_NOTRECORDED) {
194                         if (!create)
195                                 /*
196                                  * Allocated but not recorded, read treats
197                                  * this as a hole
198                                  */
199                                 goto unlock;
200 #ifdef _JFS_4K
201                         XADoffset(&xad, lblock64);
202                         XADlength(&xad, xlen);
203                         XADaddress(&xad, xaddr);
204 #else                           /* _JFS_4K */
205                         /*
206                          * As long as block size = 4K, this isn't a problem.
207                          * We should mark the whole page not ABNR, but how
208                          * will we know to mark the other blocks BH_New?
209                          */
210                         BUG();
211 #endif                          /* _JFS_4K */
212                         rc = extRecord(ip, &xad);
213                         if (rc)
214                                 goto unlock;
215                         set_buffer_new(bh_result);
216                 }
217
218                 map_bh(bh_result, ip->i_sb, xaddr);
219                 bh_result->b_size = xlen << ip->i_blkbits;
220                 goto unlock;
221         }
222         if (!create)
223                 goto unlock;
224
225         /*
226          * Allocate a new block
227          */
228 #ifdef _JFS_4K
229         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
230                 goto unlock;
231         rc = extAlloc(ip, xlen, lblock64, &xad, FALSE);
232         if (rc)
233                 goto unlock;
234
235         set_buffer_new(bh_result);
236         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
237         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
238
239 #else                           /* _JFS_4K */
240         /*
241          * We need to do whatever it takes to keep all but the last buffers
242          * in 4K pages - see jfs_write.c
243          */
244         BUG();
245 #endif                          /* _JFS_4K */
246
247       unlock:
248         /*
249          * Release lock on inode
250          */
251         if (create)
252                 IWRITE_UNLOCK(ip);
253         else
254                 IREAD_UNLOCK(ip);
255         return rc;
256 }
257
258 static int jfs_get_block(struct inode *ip, sector_t lblock,
259                          struct buffer_head *bh_result, int create)
260 {
261         return jfs_get_blocks(ip, lblock, 1, bh_result, create);
262 }
263
264 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
265 {
266         return nobh_writepage(page, jfs_get_block, wbc);
267 }
268
269 static int jfs_writepages(struct address_space *mapping,
270                         struct writeback_control *wbc)
271 {
272         return mpage_writepages(mapping, wbc, jfs_get_block);
273 }
274
275 static int jfs_readpage(struct file *file, struct page *page)
276 {
277         return mpage_readpage(page, jfs_get_block);
278 }
279
280 static int jfs_readpages(struct file *file, struct address_space *mapping,
281                 struct list_head *pages, unsigned nr_pages)
282 {
283         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
284 }
285
286 static int jfs_prepare_write(struct file *file,
287                              struct page *page, unsigned from, unsigned to)
288 {
289         return nobh_prepare_write(page, from, to, jfs_get_block);
290 }
291
292 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
293 {
294         return generic_block_bmap(mapping, block, jfs_get_block);
295 }
296
297 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
298         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
299 {
300         struct file *file = iocb->ki_filp;
301         struct inode *inode = file->f_mapping->host;
302
303         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
304                                 offset, nr_segs, jfs_get_blocks, NULL);
305 }
306
307 struct address_space_operations jfs_aops = {
308         .readpage       = jfs_readpage,
309         .readpages      = jfs_readpages,
310         .writepage      = jfs_writepage,
311         .writepages     = jfs_writepages,
312         .sync_page      = block_sync_page,
313         .prepare_write  = jfs_prepare_write,
314         .commit_write   = nobh_commit_write,
315         .bmap           = jfs_bmap,
316         .direct_IO      = jfs_direct_IO,
317 };
318
319 /*
320  * Guts of jfs_truncate.  Called with locks already held.  Can be called
321  * with directory for truncating directory index table.
322  */
323 void jfs_truncate_nolock(struct inode *ip, loff_t length)
324 {
325         loff_t newsize;
326         tid_t tid;
327
328         ASSERT(length >= 0);
329
330         if (test_cflag(COMMIT_Nolink, ip)) {
331                 xtTruncate(0, ip, length, COMMIT_WMAP);
332                 return;
333         }
334
335         do {
336                 tid = txBegin(ip->i_sb, 0);
337
338                 /*
339                  * The commit_sem cannot be taken before txBegin.
340                  * txBegin may block and there is a chance the inode
341                  * could be marked dirty and need to be committed
342                  * before txBegin unblocks
343                  */
344                 down(&JFS_IP(ip)->commit_sem);
345
346                 newsize = xtTruncate(tid, ip, length,
347                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
348                 if (newsize < 0) {
349                         txEnd(tid);
350                         up(&JFS_IP(ip)->commit_sem);
351                         break;
352                 }
353
354                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
355                 mark_inode_dirty(ip);
356
357                 txCommit(tid, 1, &ip, 0);
358                 txEnd(tid);
359                 up(&JFS_IP(ip)->commit_sem);
360         } while (newsize > length);     /* Truncate isn't always atomic */
361 }
362
363 void jfs_truncate(struct inode *ip)
364 {
365         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
366
367         nobh_truncate_page(ip->i_mapping, ip->i_size);
368
369         IWRITE_LOCK(ip);
370         jfs_truncate_nolock(ip, ip->i_size);
371         IWRITE_UNLOCK(ip);
372 }