Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6.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 <linux/writeback.h>
26 #include "jfs_incore.h"
27 #include "jfs_inode.h"
28 #include "jfs_filsys.h"
29 #include "jfs_imap.h"
30 #include "jfs_extent.h"
31 #include "jfs_unicode.h"
32 #include "jfs_debug.h"
33
34
35 struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
36 {
37         struct inode *inode;
38         int ret;
39
40         inode = iget_locked(sb, ino);
41         if (!inode)
42                 return ERR_PTR(-ENOMEM);
43         if (!(inode->i_state & I_NEW))
44                 return inode;
45
46         ret = diRead(inode);
47         if (ret < 0) {
48                 iget_failed(inode);
49                 return ERR_PTR(ret);
50         }
51
52         if (S_ISREG(inode->i_mode)) {
53                 inode->i_op = &jfs_file_inode_operations;
54                 inode->i_fop = &jfs_file_operations;
55                 inode->i_mapping->a_ops = &jfs_aops;
56         } else if (S_ISDIR(inode->i_mode)) {
57                 inode->i_op = &jfs_dir_inode_operations;
58                 inode->i_fop = &jfs_dir_operations;
59         } else if (S_ISLNK(inode->i_mode)) {
60                 if (inode->i_size >= IDATASIZE) {
61                         inode->i_op = &page_symlink_inode_operations;
62                         inode->i_mapping->a_ops = &jfs_aops;
63                 } else {
64                         inode->i_op = &jfs_symlink_inode_operations;
65                         /*
66                          * The inline data should be null-terminated, but
67                          * don't let on-disk corruption crash the kernel
68                          */
69                         JFS_IP(inode)->i_inline[inode->i_size] = '\0';
70                 }
71         } else {
72                 inode->i_op = &jfs_file_inode_operations;
73                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
74         }
75         unlock_new_inode(inode);
76         return inode;
77 }
78
79 /*
80  * Workhorse of both fsync & write_inode
81  */
82 int jfs_commit_inode(struct inode *inode, int wait)
83 {
84         int rc = 0;
85         tid_t tid;
86         static int noisy = 5;
87
88         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
89
90         /*
91          * Don't commit if inode has been committed since last being
92          * marked dirty, or if it has been deleted.
93          */
94         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
95                 return 0;
96
97         if (isReadOnly(inode)) {
98                 /* kernel allows writes to devices on read-only
99                  * partitions and may think inode is dirty
100                  */
101                 if (!special_file(inode->i_mode) && noisy) {
102                         jfs_err("jfs_commit_inode(0x%p) called on "
103                                    "read-only volume", inode);
104                         jfs_err("Is remount racy?");
105                         noisy--;
106                 }
107                 return 0;
108         }
109
110         tid = txBegin(inode->i_sb, COMMIT_INODE);
111         mutex_lock(&JFS_IP(inode)->commit_mutex);
112
113         /*
114          * Retest inode state after taking commit_mutex
115          */
116         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
117                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
118
119         txEnd(tid);
120         mutex_unlock(&JFS_IP(inode)->commit_mutex);
121         return rc;
122 }
123
124 int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
125 {
126         int wait = wbc->sync_mode == WB_SYNC_ALL;
127
128         if (test_cflag(COMMIT_Nolink, inode))
129                 return 0;
130         /*
131          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
132          * It has been committed since the last change, but was still
133          * on the dirty inode list.
134          */
135          if (!test_cflag(COMMIT_Dirty, inode)) {
136                 /* Make sure committed changes hit the disk */
137                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
138                 return 0;
139          }
140
141         if (jfs_commit_inode(inode, wait)) {
142                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
143                 return -EIO;
144         } else
145                 return 0;
146 }
147
148 void jfs_delete_inode(struct inode *inode)
149 {
150         jfs_info("In jfs_delete_inode, inode = 0x%p", inode);
151
152         if (!is_bad_inode(inode))
153                 dquot_initialize(inode);
154
155         if (!is_bad_inode(inode) &&
156             (JFS_IP(inode)->fileset == FILESYSTEM_I)) {
157                 truncate_inode_pages(&inode->i_data, 0);
158
159                 if (test_cflag(COMMIT_Freewmap, inode))
160                         jfs_free_zero_link(inode);
161
162                 diFree(inode);
163
164                 /*
165                  * Free the inode from the quota allocation.
166                  */
167                 dquot_initialize(inode);
168                 dquot_free_inode(inode);
169                 dquot_drop(inode);
170         }
171
172         clear_inode(inode);
173 }
174
175 void jfs_dirty_inode(struct inode *inode)
176 {
177         static int noisy = 5;
178
179         if (isReadOnly(inode)) {
180                 if (!special_file(inode->i_mode) && noisy) {
181                         /* kernel allows writes to devices on read-only
182                          * partitions and may try to mark inode dirty
183                          */
184                         jfs_err("jfs_dirty_inode called on read-only volume");
185                         jfs_err("Is remount racy?");
186                         noisy--;
187                 }
188                 return;
189         }
190
191         set_cflag(COMMIT_Dirty, inode);
192 }
193
194 int jfs_get_block(struct inode *ip, sector_t lblock,
195                   struct buffer_head *bh_result, int create)
196 {
197         s64 lblock64 = lblock;
198         int rc = 0;
199         xad_t xad;
200         s64 xaddr;
201         int xflag;
202         s32 xlen = bh_result->b_size >> ip->i_blkbits;
203
204         /*
205          * Take appropriate lock on inode
206          */
207         if (create)
208                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
209         else
210                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
211
212         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
213             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
214             xaddr) {
215                 if (xflag & XAD_NOTRECORDED) {
216                         if (!create)
217                                 /*
218                                  * Allocated but not recorded, read treats
219                                  * this as a hole
220                                  */
221                                 goto unlock;
222 #ifdef _JFS_4K
223                         XADoffset(&xad, lblock64);
224                         XADlength(&xad, xlen);
225                         XADaddress(&xad, xaddr);
226 #else                           /* _JFS_4K */
227                         /*
228                          * As long as block size = 4K, this isn't a problem.
229                          * We should mark the whole page not ABNR, but how
230                          * will we know to mark the other blocks BH_New?
231                          */
232                         BUG();
233 #endif                          /* _JFS_4K */
234                         rc = extRecord(ip, &xad);
235                         if (rc)
236                                 goto unlock;
237                         set_buffer_new(bh_result);
238                 }
239
240                 map_bh(bh_result, ip->i_sb, xaddr);
241                 bh_result->b_size = xlen << ip->i_blkbits;
242                 goto unlock;
243         }
244         if (!create)
245                 goto unlock;
246
247         /*
248          * Allocate a new block
249          */
250 #ifdef _JFS_4K
251         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
252                 goto unlock;
253         rc = extAlloc(ip, xlen, lblock64, &xad, false);
254         if (rc)
255                 goto unlock;
256
257         set_buffer_new(bh_result);
258         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
259         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
260
261 #else                           /* _JFS_4K */
262         /*
263          * We need to do whatever it takes to keep all but the last buffers
264          * in 4K pages - see jfs_write.c
265          */
266         BUG();
267 #endif                          /* _JFS_4K */
268
269       unlock:
270         /*
271          * Release lock on inode
272          */
273         if (create)
274                 IWRITE_UNLOCK(ip);
275         else
276                 IREAD_UNLOCK(ip);
277         return rc;
278 }
279
280 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
281 {
282         return block_write_full_page(page, jfs_get_block, wbc);
283 }
284
285 static int jfs_writepages(struct address_space *mapping,
286                         struct writeback_control *wbc)
287 {
288         return mpage_writepages(mapping, wbc, jfs_get_block);
289 }
290
291 static int jfs_readpage(struct file *file, struct page *page)
292 {
293         return mpage_readpage(page, jfs_get_block);
294 }
295
296 static int jfs_readpages(struct file *file, struct address_space *mapping,
297                 struct list_head *pages, unsigned nr_pages)
298 {
299         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
300 }
301
302 static int jfs_write_begin(struct file *file, struct address_space *mapping,
303                                 loff_t pos, unsigned len, unsigned flags,
304                                 struct page **pagep, void **fsdata)
305 {
306         return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
307                                 jfs_get_block);
308 }
309
310 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
311 {
312         return generic_block_bmap(mapping, block, jfs_get_block);
313 }
314
315 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
316         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
317 {
318         struct file *file = iocb->ki_filp;
319         struct inode *inode = file->f_mapping->host;
320
321         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
322                                 offset, nr_segs, jfs_get_block, NULL);
323 }
324
325 const struct address_space_operations jfs_aops = {
326         .readpage       = jfs_readpage,
327         .readpages      = jfs_readpages,
328         .writepage      = jfs_writepage,
329         .writepages     = jfs_writepages,
330         .sync_page      = block_sync_page,
331         .write_begin    = jfs_write_begin,
332         .write_end      = nobh_write_end,
333         .bmap           = jfs_bmap,
334         .direct_IO      = jfs_direct_IO,
335 };
336
337 /*
338  * Guts of jfs_truncate.  Called with locks already held.  Can be called
339  * with directory for truncating directory index table.
340  */
341 void jfs_truncate_nolock(struct inode *ip, loff_t length)
342 {
343         loff_t newsize;
344         tid_t tid;
345
346         ASSERT(length >= 0);
347
348         if (test_cflag(COMMIT_Nolink, ip)) {
349                 xtTruncate(0, ip, length, COMMIT_WMAP);
350                 return;
351         }
352
353         do {
354                 tid = txBegin(ip->i_sb, 0);
355
356                 /*
357                  * The commit_mutex cannot be taken before txBegin.
358                  * txBegin may block and there is a chance the inode
359                  * could be marked dirty and need to be committed
360                  * before txBegin unblocks
361                  */
362                 mutex_lock(&JFS_IP(ip)->commit_mutex);
363
364                 newsize = xtTruncate(tid, ip, length,
365                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
366                 if (newsize < 0) {
367                         txEnd(tid);
368                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
369                         break;
370                 }
371
372                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
373                 mark_inode_dirty(ip);
374
375                 txCommit(tid, 1, &ip, 0);
376                 txEnd(tid);
377                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
378         } while (newsize > length);     /* Truncate isn't always atomic */
379 }
380
381 void jfs_truncate(struct inode *ip)
382 {
383         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
384
385         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
386
387         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
388         jfs_truncate_nolock(ip, ip->i_size);
389         IWRITE_UNLOCK(ip);
390 }