Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm
[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         } else {
65                 inode->i_op = &jfs_file_inode_operations;
66                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
67         }
68         unlock_new_inode(inode);
69         return inode;
70 }
71
72 /*
73  * Workhorse of both fsync & write_inode
74  */
75 int jfs_commit_inode(struct inode *inode, int wait)
76 {
77         int rc = 0;
78         tid_t tid;
79         static int noisy = 5;
80
81         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
82
83         /*
84          * Don't commit if inode has been committed since last being
85          * marked dirty, or if it has been deleted.
86          */
87         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
88                 return 0;
89
90         if (isReadOnly(inode)) {
91                 /* kernel allows writes to devices on read-only
92                  * partitions and may think inode is dirty
93                  */
94                 if (!special_file(inode->i_mode) && noisy) {
95                         jfs_err("jfs_commit_inode(0x%p) called on "
96                                    "read-only volume", inode);
97                         jfs_err("Is remount racy?");
98                         noisy--;
99                 }
100                 return 0;
101         }
102
103         tid = txBegin(inode->i_sb, COMMIT_INODE);
104         mutex_lock(&JFS_IP(inode)->commit_mutex);
105
106         /*
107          * Retest inode state after taking commit_mutex
108          */
109         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
110                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
111
112         txEnd(tid);
113         mutex_unlock(&JFS_IP(inode)->commit_mutex);
114         return rc;
115 }
116
117 int jfs_write_inode(struct inode *inode, int wait)
118 {
119         if (test_cflag(COMMIT_Nolink, inode))
120                 return 0;
121         /*
122          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
123          * It has been committed since the last change, but was still
124          * on the dirty inode list.
125          */
126          if (!test_cflag(COMMIT_Dirty, inode)) {
127                 /* Make sure committed changes hit the disk */
128                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
129                 return 0;
130          }
131
132         if (jfs_commit_inode(inode, wait)) {
133                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
134                 return -EIO;
135         } else
136                 return 0;
137 }
138
139 void jfs_delete_inode(struct inode *inode)
140 {
141         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
142
143         if (!is_bad_inode(inode) &&
144             (JFS_IP(inode)->fileset == FILESYSTEM_I)) {
145                 truncate_inode_pages(&inode->i_data, 0);
146
147                 if (test_cflag(COMMIT_Freewmap, inode))
148                         jfs_free_zero_link(inode);
149
150                 diFree(inode);
151
152                 /*
153                  * Free the inode from the quota allocation.
154                  */
155                 DQUOT_INIT(inode);
156                 DQUOT_FREE_INODE(inode);
157                 DQUOT_DROP(inode);
158         }
159
160         clear_inode(inode);
161 }
162
163 void jfs_dirty_inode(struct inode *inode)
164 {
165         static int noisy = 5;
166
167         if (isReadOnly(inode)) {
168                 if (!special_file(inode->i_mode) && noisy) {
169                         /* kernel allows writes to devices on read-only
170                          * partitions and may try to mark inode dirty
171                          */
172                         jfs_err("jfs_dirty_inode called on read-only volume");
173                         jfs_err("Is remount racy?");
174                         noisy--;
175                 }
176                 return;
177         }
178
179         set_cflag(COMMIT_Dirty, inode);
180 }
181
182 int jfs_get_block(struct inode *ip, sector_t lblock,
183                   struct buffer_head *bh_result, int create)
184 {
185         s64 lblock64 = lblock;
186         int rc = 0;
187         xad_t xad;
188         s64 xaddr;
189         int xflag;
190         s32 xlen = bh_result->b_size >> ip->i_blkbits;
191
192         /*
193          * Take appropriate lock on inode
194          */
195         if (create)
196                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
197         else
198                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
199
200         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
201             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
202             xaddr) {
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, xlen, 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 (create)
262                 IWRITE_UNLOCK(ip);
263         else
264                 IREAD_UNLOCK(ip);
265         return rc;
266 }
267
268 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
269 {
270         return block_write_full_page(page, jfs_get_block, wbc);
271 }
272
273 static int jfs_writepages(struct address_space *mapping,
274                         struct writeback_control *wbc)
275 {
276         return mpage_writepages(mapping, wbc, jfs_get_block);
277 }
278
279 static int jfs_readpage(struct file *file, struct page *page)
280 {
281         return mpage_readpage(page, jfs_get_block);
282 }
283
284 static int jfs_readpages(struct file *file, struct address_space *mapping,
285                 struct list_head *pages, unsigned nr_pages)
286 {
287         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
288 }
289
290 static int jfs_write_begin(struct file *file, struct address_space *mapping,
291                                 loff_t pos, unsigned len, unsigned flags,
292                                 struct page **pagep, void **fsdata)
293 {
294         return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
295                                 jfs_get_block);
296 }
297
298 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
299 {
300         return generic_block_bmap(mapping, block, jfs_get_block);
301 }
302
303 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
304         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
305 {
306         struct file *file = iocb->ki_filp;
307         struct inode *inode = file->f_mapping->host;
308
309         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
310                                 offset, nr_segs, jfs_get_block, NULL);
311 }
312
313 const struct address_space_operations jfs_aops = {
314         .readpage       = jfs_readpage,
315         .readpages      = jfs_readpages,
316         .writepage      = jfs_writepage,
317         .writepages     = jfs_writepages,
318         .sync_page      = block_sync_page,
319         .write_begin    = jfs_write_begin,
320         .write_end      = nobh_write_end,
321         .bmap           = jfs_bmap,
322         .direct_IO      = jfs_direct_IO,
323 };
324
325 /*
326  * Guts of jfs_truncate.  Called with locks already held.  Can be called
327  * with directory for truncating directory index table.
328  */
329 void jfs_truncate_nolock(struct inode *ip, loff_t length)
330 {
331         loff_t newsize;
332         tid_t tid;
333
334         ASSERT(length >= 0);
335
336         if (test_cflag(COMMIT_Nolink, ip)) {
337                 xtTruncate(0, ip, length, COMMIT_WMAP);
338                 return;
339         }
340
341         do {
342                 tid = txBegin(ip->i_sb, 0);
343
344                 /*
345                  * The commit_mutex cannot be taken before txBegin.
346                  * txBegin may block and there is a chance the inode
347                  * could be marked dirty and need to be committed
348                  * before txBegin unblocks
349                  */
350                 mutex_lock(&JFS_IP(ip)->commit_mutex);
351
352                 newsize = xtTruncate(tid, ip, length,
353                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
354                 if (newsize < 0) {
355                         txEnd(tid);
356                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
357                         break;
358                 }
359
360                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
361                 mark_inode_dirty(ip);
362
363                 txCommit(tid, 1, &ip, 0);
364                 txEnd(tid);
365                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
366         } while (newsize > length);     /* Truncate isn't always atomic */
367 }
368
369 void jfs_truncate(struct inode *ip)
370 {
371         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
372
373         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
374
375         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
376         jfs_truncate_nolock(ip, ip->i_size);
377         IWRITE_UNLOCK(ip);
378 }