Linux 2.6.31-rc6
[linux-2.6] / 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 struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
35 {
36         struct inode *inode;
37         int ret;
38
39         inode = iget_locked(sb, ino);
40         if (!inode)
41                 return ERR_PTR(-ENOMEM);
42         if (!(inode->i_state & I_NEW))
43                 return inode;
44
45         ret = diRead(inode);
46         if (ret < 0) {
47                 iget_failed(inode);
48                 return ERR_PTR(ret);
49         }
50
51         if (S_ISREG(inode->i_mode)) {
52                 inode->i_op = &jfs_file_inode_operations;
53                 inode->i_fop = &jfs_file_operations;
54                 inode->i_mapping->a_ops = &jfs_aops;
55         } else if (S_ISDIR(inode->i_mode)) {
56                 inode->i_op = &jfs_dir_inode_operations;
57                 inode->i_fop = &jfs_dir_operations;
58         } else if (S_ISLNK(inode->i_mode)) {
59                 if (inode->i_size >= IDATASIZE) {
60                         inode->i_op = &page_symlink_inode_operations;
61                         inode->i_mapping->a_ops = &jfs_aops;
62                 } else {
63                         inode->i_op = &jfs_symlink_inode_operations;
64                         /*
65                          * The inline data should be null-terminated, but
66                          * don't let on-disk corruption crash the kernel
67                          */
68                         JFS_IP(inode)->i_inline[inode->i_size] = '\0';
69                 }
70         } else {
71                 inode->i_op = &jfs_file_inode_operations;
72                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
73         }
74         unlock_new_inode(inode);
75         return inode;
76 }
77
78 /*
79  * Workhorse of both fsync & write_inode
80  */
81 int jfs_commit_inode(struct inode *inode, int wait)
82 {
83         int rc = 0;
84         tid_t tid;
85         static int noisy = 5;
86
87         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
88
89         /*
90          * Don't commit if inode has been committed since last being
91          * marked dirty, or if it has been deleted.
92          */
93         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
94                 return 0;
95
96         if (isReadOnly(inode)) {
97                 /* kernel allows writes to devices on read-only
98                  * partitions and may think inode is dirty
99                  */
100                 if (!special_file(inode->i_mode) && noisy) {
101                         jfs_err("jfs_commit_inode(0x%p) called on "
102                                    "read-only volume", inode);
103                         jfs_err("Is remount racy?");
104                         noisy--;
105                 }
106                 return 0;
107         }
108
109         tid = txBegin(inode->i_sb, COMMIT_INODE);
110         mutex_lock(&JFS_IP(inode)->commit_mutex);
111
112         /*
113          * Retest inode state after taking commit_mutex
114          */
115         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
116                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
117
118         txEnd(tid);
119         mutex_unlock(&JFS_IP(inode)->commit_mutex);
120         return rc;
121 }
122
123 int jfs_write_inode(struct inode *inode, int wait)
124 {
125         if (test_cflag(COMMIT_Nolink, inode))
126                 return 0;
127         /*
128          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
129          * It has been committed since the last change, but was still
130          * on the dirty inode list.
131          */
132          if (!test_cflag(COMMIT_Dirty, inode)) {
133                 /* Make sure committed changes hit the disk */
134                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
135                 return 0;
136          }
137
138         if (jfs_commit_inode(inode, wait)) {
139                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
140                 return -EIO;
141         } else
142                 return 0;
143 }
144
145 void jfs_delete_inode(struct inode *inode)
146 {
147         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
148
149         if (!is_bad_inode(inode) &&
150             (JFS_IP(inode)->fileset == FILESYSTEM_I)) {
151                 truncate_inode_pages(&inode->i_data, 0);
152
153                 if (test_cflag(COMMIT_Freewmap, inode))
154                         jfs_free_zero_link(inode);
155
156                 diFree(inode);
157
158                 /*
159                  * Free the inode from the quota allocation.
160                  */
161                 vfs_dq_init(inode);
162                 vfs_dq_free_inode(inode);
163                 vfs_dq_drop(inode);
164         }
165
166         clear_inode(inode);
167 }
168
169 void jfs_dirty_inode(struct inode *inode)
170 {
171         static int noisy = 5;
172
173         if (isReadOnly(inode)) {
174                 if (!special_file(inode->i_mode) && noisy) {
175                         /* kernel allows writes to devices on read-only
176                          * partitions and may try to mark inode dirty
177                          */
178                         jfs_err("jfs_dirty_inode called on read-only volume");
179                         jfs_err("Is remount racy?");
180                         noisy--;
181                 }
182                 return;
183         }
184
185         set_cflag(COMMIT_Dirty, inode);
186 }
187
188 int jfs_get_block(struct inode *ip, sector_t lblock,
189                   struct buffer_head *bh_result, int create)
190 {
191         s64 lblock64 = lblock;
192         int rc = 0;
193         xad_t xad;
194         s64 xaddr;
195         int xflag;
196         s32 xlen = bh_result->b_size >> ip->i_blkbits;
197
198         /*
199          * Take appropriate lock on inode
200          */
201         if (create)
202                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
203         else
204                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
205
206         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
207             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
208             xaddr) {
209                 if (xflag & XAD_NOTRECORDED) {
210                         if (!create)
211                                 /*
212                                  * Allocated but not recorded, read treats
213                                  * this as a hole
214                                  */
215                                 goto unlock;
216 #ifdef _JFS_4K
217                         XADoffset(&xad, lblock64);
218                         XADlength(&xad, xlen);
219                         XADaddress(&xad, xaddr);
220 #else                           /* _JFS_4K */
221                         /*
222                          * As long as block size = 4K, this isn't a problem.
223                          * We should mark the whole page not ABNR, but how
224                          * will we know to mark the other blocks BH_New?
225                          */
226                         BUG();
227 #endif                          /* _JFS_4K */
228                         rc = extRecord(ip, &xad);
229                         if (rc)
230                                 goto unlock;
231                         set_buffer_new(bh_result);
232                 }
233
234                 map_bh(bh_result, ip->i_sb, xaddr);
235                 bh_result->b_size = xlen << ip->i_blkbits;
236                 goto unlock;
237         }
238         if (!create)
239                 goto unlock;
240
241         /*
242          * Allocate a new block
243          */
244 #ifdef _JFS_4K
245         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
246                 goto unlock;
247         rc = extAlloc(ip, xlen, lblock64, &xad, false);
248         if (rc)
249                 goto unlock;
250
251         set_buffer_new(bh_result);
252         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
253         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
254
255 #else                           /* _JFS_4K */
256         /*
257          * We need to do whatever it takes to keep all but the last buffers
258          * in 4K pages - see jfs_write.c
259          */
260         BUG();
261 #endif                          /* _JFS_4K */
262
263       unlock:
264         /*
265          * Release lock on inode
266          */
267         if (create)
268                 IWRITE_UNLOCK(ip);
269         else
270                 IREAD_UNLOCK(ip);
271         return rc;
272 }
273
274 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
275 {
276         return block_write_full_page(page, jfs_get_block, wbc);
277 }
278
279 static int jfs_writepages(struct address_space *mapping,
280                         struct writeback_control *wbc)
281 {
282         return mpage_writepages(mapping, wbc, jfs_get_block);
283 }
284
285 static int jfs_readpage(struct file *file, struct page *page)
286 {
287         return mpage_readpage(page, jfs_get_block);
288 }
289
290 static int jfs_readpages(struct file *file, struct address_space *mapping,
291                 struct list_head *pages, unsigned nr_pages)
292 {
293         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
294 }
295
296 static int jfs_write_begin(struct file *file, struct address_space *mapping,
297                                 loff_t pos, unsigned len, unsigned flags,
298                                 struct page **pagep, void **fsdata)
299 {
300         return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
301                                 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_block, NULL);
317 }
318
319 const 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         .write_begin    = jfs_write_begin,
326         .write_end      = nobh_write_end,
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_mutex 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                 mutex_lock(&JFS_IP(ip)->commit_mutex);
357
358                 newsize = xtTruncate(tid, ip, length,
359                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
360                 if (newsize < 0) {
361                         txEnd(tid);
362                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
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                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
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, jfs_get_block);
380
381         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
382         jfs_truncate_nolock(ip, ip->i_size);
383         IWRITE_UNLOCK(ip);
384 }