Linux-2.6.12-rc2
[linux-3.10.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_filsys.h"
27 #include "jfs_imap.h"
28 #include "jfs_extent.h"
29 #include "jfs_unicode.h"
30 #include "jfs_debug.h"
31
32
33 extern struct inode_operations jfs_dir_inode_operations;
34 extern struct inode_operations jfs_file_inode_operations;
35 extern struct inode_operations jfs_symlink_inode_operations;
36 extern struct file_operations jfs_dir_operations;
37 extern struct file_operations jfs_file_operations;
38 struct address_space_operations jfs_aops;
39 extern int freeZeroLink(struct inode *);
40
41 void jfs_read_inode(struct inode *inode)
42 {
43         if (diRead(inode)) { 
44                 make_bad_inode(inode);
45                 return;
46         }
47
48         if (S_ISREG(inode->i_mode)) {
49                 inode->i_op = &jfs_file_inode_operations;
50                 inode->i_fop = &jfs_file_operations;
51                 inode->i_mapping->a_ops = &jfs_aops;
52         } else if (S_ISDIR(inode->i_mode)) {
53                 inode->i_op = &jfs_dir_inode_operations;
54                 inode->i_fop = &jfs_dir_operations;
55         } else if (S_ISLNK(inode->i_mode)) {
56                 if (inode->i_size >= IDATASIZE) {
57                         inode->i_op = &page_symlink_inode_operations;
58                         inode->i_mapping->a_ops = &jfs_aops;
59                 } else
60                         inode->i_op = &jfs_symlink_inode_operations;
61         } else {
62                 inode->i_op = &jfs_file_inode_operations;
63                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
64         }
65 }
66
67 /*
68  * Workhorse of both fsync & write_inode
69  */
70 int jfs_commit_inode(struct inode *inode, int wait)
71 {
72         int rc = 0;
73         tid_t tid;
74         static int noisy = 5;
75
76         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
77
78         /*
79          * Don't commit if inode has been committed since last being
80          * marked dirty, or if it has been deleted.
81          */
82         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
83                 return 0;
84
85         if (isReadOnly(inode)) {
86                 /* kernel allows writes to devices on read-only
87                  * partitions and may think inode is dirty
88                  */
89                 if (!special_file(inode->i_mode) && noisy) {
90                         jfs_err("jfs_commit_inode(0x%p) called on "
91                                    "read-only volume", inode);
92                         jfs_err("Is remount racy?");
93                         noisy--;
94                 }
95                 return 0;
96         }
97
98         tid = txBegin(inode->i_sb, COMMIT_INODE);
99         down(&JFS_IP(inode)->commit_sem);
100
101         /*
102          * Retest inode state after taking commit_sem
103          */
104         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
105                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
106
107         txEnd(tid);
108         up(&JFS_IP(inode)->commit_sem);
109         return rc;
110 }
111
112 int jfs_write_inode(struct inode *inode, int wait)
113 {
114         if (test_cflag(COMMIT_Nolink, inode))
115                 return 0;
116         /*
117          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
118          * It has been committed since the last change, but was still
119          * on the dirty inode list.
120          */
121          if (!test_cflag(COMMIT_Dirty, inode)) {
122                 /* Make sure committed changes hit the disk */
123                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
124                 return 0;
125          }
126
127         if (jfs_commit_inode(inode, wait)) {
128                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
129                 return -EIO;
130         } else
131                 return 0;
132 }
133
134 void jfs_delete_inode(struct inode *inode)
135 {
136         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
137
138         if (test_cflag(COMMIT_Freewmap, inode))
139                 freeZeroLink(inode);
140
141         diFree(inode);
142
143         /*
144          * Free the inode from the quota allocation.
145          */
146         DQUOT_INIT(inode);
147         DQUOT_FREE_INODE(inode);
148         DQUOT_DROP(inode);
149
150         clear_inode(inode);
151 }
152
153 void jfs_dirty_inode(struct inode *inode)
154 {
155         static int noisy = 5;
156
157         if (isReadOnly(inode)) {
158                 if (!special_file(inode->i_mode) && noisy) {
159                         /* kernel allows writes to devices on read-only
160                          * partitions and may try to mark inode dirty
161                          */
162                         jfs_err("jfs_dirty_inode called on read-only volume");
163                         jfs_err("Is remount racy?");
164                         noisy--;
165                 }
166                 return;
167         }
168
169         set_cflag(COMMIT_Dirty, inode);
170 }
171
172 static int
173 jfs_get_blocks(struct inode *ip, sector_t lblock, unsigned long max_blocks,
174                         struct buffer_head *bh_result, int create)
175 {
176         s64 lblock64 = lblock;
177         int rc = 0;
178         int take_locks;
179         xad_t xad;
180         s64 xaddr;
181         int xflag;
182         s32 xlen;
183
184         /*
185          * If this is a special inode (imap, dmap)
186          * the lock should already be taken
187          */
188         take_locks = (JFS_IP(ip)->fileset != AGGREGATE_I);
189
190         /*
191          * Take appropriate lock on inode
192          */
193         if (take_locks) {
194                 if (create)
195                         IWRITE_LOCK(ip);
196                 else
197                         IREAD_LOCK(ip);
198         }
199
200         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
201             (xtLookup(ip, lblock64, max_blocks, &xflag, &xaddr, &xlen, 0)
202              == 0) && xlen) {
203                 if (xflag & XAD_NOTRECORDED) {
204                         if (!create)
205                                 /*
206                                  * Allocated but not recorded, read treats
207                                  * this as a hole
208                                  */
209                                 goto unlock;
210 #ifdef _JFS_4K
211                         XADoffset(&xad, lblock64);
212                         XADlength(&xad, xlen);
213                         XADaddress(&xad, xaddr);
214 #else                           /* _JFS_4K */
215                         /*
216                          * As long as block size = 4K, this isn't a problem.
217                          * We should mark the whole page not ABNR, but how
218                          * will we know to mark the other blocks BH_New?
219                          */
220                         BUG();
221 #endif                          /* _JFS_4K */
222                         rc = extRecord(ip, &xad);
223                         if (rc)
224                                 goto unlock;
225                         set_buffer_new(bh_result);
226                 }
227
228                 map_bh(bh_result, ip->i_sb, xaddr);
229                 bh_result->b_size = xlen << ip->i_blkbits;
230                 goto unlock;
231         }
232         if (!create)
233                 goto unlock;
234
235         /*
236          * Allocate a new block
237          */
238 #ifdef _JFS_4K
239         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
240                 goto unlock;
241         rc = extAlloc(ip, max_blocks, lblock64, &xad, FALSE);
242         if (rc)
243                 goto unlock;
244
245         set_buffer_new(bh_result);
246         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
247         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
248
249 #else                           /* _JFS_4K */
250         /*
251          * We need to do whatever it takes to keep all but the last buffers
252          * in 4K pages - see jfs_write.c
253          */
254         BUG();
255 #endif                          /* _JFS_4K */
256
257       unlock:
258         /*
259          * Release lock on inode
260          */
261         if (take_locks) {
262                 if (create)
263                         IWRITE_UNLOCK(ip);
264                 else
265                         IREAD_UNLOCK(ip);
266         }
267         return rc;
268 }
269
270 static int jfs_get_block(struct inode *ip, sector_t lblock,
271                          struct buffer_head *bh_result, int create)
272 {
273         return jfs_get_blocks(ip, lblock, 1, bh_result, create);
274 }
275
276 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
277 {
278         return nobh_writepage(page, jfs_get_block, wbc);
279 }
280
281 static int jfs_writepages(struct address_space *mapping,
282                         struct writeback_control *wbc)
283 {
284         return mpage_writepages(mapping, wbc, jfs_get_block);
285 }
286
287 static int jfs_readpage(struct file *file, struct page *page)
288 {
289         return mpage_readpage(page, jfs_get_block);
290 }
291
292 static int jfs_readpages(struct file *file, struct address_space *mapping,
293                 struct list_head *pages, unsigned nr_pages)
294 {
295         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
296 }
297
298 static int jfs_prepare_write(struct file *file,
299                              struct page *page, unsigned from, unsigned to)
300 {
301         return nobh_prepare_write(page, from, to, jfs_get_block);
302 }
303
304 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
305 {
306         return generic_block_bmap(mapping, block, jfs_get_block);
307 }
308
309 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
310         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
311 {
312         struct file *file = iocb->ki_filp;
313         struct inode *inode = file->f_mapping->host;
314
315         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
316                                 offset, nr_segs, jfs_get_blocks, NULL);
317 }
318
319 struct address_space_operations jfs_aops = {
320         .readpage       = jfs_readpage,
321         .readpages      = jfs_readpages,
322         .writepage      = jfs_writepage,
323         .writepages     = jfs_writepages,
324         .sync_page      = block_sync_page,
325         .prepare_write  = jfs_prepare_write,
326         .commit_write   = nobh_commit_write,
327         .bmap           = jfs_bmap,
328         .direct_IO      = jfs_direct_IO,
329 };
330
331 /*
332  * Guts of jfs_truncate.  Called with locks already held.  Can be called
333  * with directory for truncating directory index table.
334  */
335 void jfs_truncate_nolock(struct inode *ip, loff_t length)
336 {
337         loff_t newsize;
338         tid_t tid;
339
340         ASSERT(length >= 0);
341
342         if (test_cflag(COMMIT_Nolink, ip)) {
343                 xtTruncate(0, ip, length, COMMIT_WMAP);
344                 return;
345         }
346
347         do {
348                 tid = txBegin(ip->i_sb, 0);
349
350                 /*
351                  * The commit_sem cannot be taken before txBegin.
352                  * txBegin may block and there is a chance the inode
353                  * could be marked dirty and need to be committed
354                  * before txBegin unblocks
355                  */
356                 down(&JFS_IP(ip)->commit_sem);
357
358                 newsize = xtTruncate(tid, ip, length,
359                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
360                 if (newsize < 0) {
361                         txEnd(tid);
362                         up(&JFS_IP(ip)->commit_sem);
363                         break;
364                 }
365
366                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
367                 mark_inode_dirty(ip);
368
369                 txCommit(tid, 1, &ip, 0);
370                 txEnd(tid);
371                 up(&JFS_IP(ip)->commit_sem);
372         } while (newsize > length);     /* Truncate isn't always atomic */
373 }
374
375 void jfs_truncate(struct inode *ip)
376 {
377         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
378
379         nobh_truncate_page(ip->i_mapping, ip->i_size);
380
381         IWRITE_LOCK(ip);
382         jfs_truncate_nolock(ip, ip->i_size);
383         IWRITE_UNLOCK(ip);
384 }