Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm
[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 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         mutex_lock(&JFS_IP(inode)->commit_mutex);
93
94         /*
95          * Retest inode state after taking commit_mutex
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         mutex_unlock(&JFS_IP(inode)->commit_mutex);
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 == FILESYSTEM_I)) {
133                 truncate_inode_pages(&inode->i_data, 0);
134
135                 if (test_cflag(COMMIT_Freewmap, inode))
136                         jfs_free_zero_link(inode);
137
138                 diFree(inode);
139
140                 /*
141                  * Free the inode from the quota allocation.
142                  */
143                 DQUOT_INIT(inode);
144                 DQUOT_FREE_INODE(inode);
145                 DQUOT_DROP(inode);
146         }
147
148         clear_inode(inode);
149 }
150
151 void jfs_dirty_inode(struct inode *inode)
152 {
153         static int noisy = 5;
154
155         if (isReadOnly(inode)) {
156                 if (!special_file(inode->i_mode) && noisy) {
157                         /* kernel allows writes to devices on read-only
158                          * partitions and may try to mark inode dirty
159                          */
160                         jfs_err("jfs_dirty_inode called on read-only volume");
161                         jfs_err("Is remount racy?");
162                         noisy--;
163                 }
164                 return;
165         }
166
167         set_cflag(COMMIT_Dirty, inode);
168 }
169
170 int jfs_get_block(struct inode *ip, sector_t lblock,
171                   struct buffer_head *bh_result, int create)
172 {
173         s64 lblock64 = lblock;
174         int rc = 0;
175         xad_t xad;
176         s64 xaddr;
177         int xflag;
178         s32 xlen = bh_result->b_size >> ip->i_blkbits;
179
180         /*
181          * Take appropriate lock on inode
182          */
183         if (create)
184                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
185         else
186                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
187
188         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
189             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
190             xaddr) {
191                 if (xflag & XAD_NOTRECORDED) {
192                         if (!create)
193                                 /*
194                                  * Allocated but not recorded, read treats
195                                  * this as a hole
196                                  */
197                                 goto unlock;
198 #ifdef _JFS_4K
199                         XADoffset(&xad, lblock64);
200                         XADlength(&xad, xlen);
201                         XADaddress(&xad, xaddr);
202 #else                           /* _JFS_4K */
203                         /*
204                          * As long as block size = 4K, this isn't a problem.
205                          * We should mark the whole page not ABNR, but how
206                          * will we know to mark the other blocks BH_New?
207                          */
208                         BUG();
209 #endif                          /* _JFS_4K */
210                         rc = extRecord(ip, &xad);
211                         if (rc)
212                                 goto unlock;
213                         set_buffer_new(bh_result);
214                 }
215
216                 map_bh(bh_result, ip->i_sb, xaddr);
217                 bh_result->b_size = xlen << ip->i_blkbits;
218                 goto unlock;
219         }
220         if (!create)
221                 goto unlock;
222
223         /*
224          * Allocate a new block
225          */
226 #ifdef _JFS_4K
227         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
228                 goto unlock;
229         rc = extAlloc(ip, xlen, lblock64, &xad, false);
230         if (rc)
231                 goto unlock;
232
233         set_buffer_new(bh_result);
234         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
235         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
236
237 #else                           /* _JFS_4K */
238         /*
239          * We need to do whatever it takes to keep all but the last buffers
240          * in 4K pages - see jfs_write.c
241          */
242         BUG();
243 #endif                          /* _JFS_4K */
244
245       unlock:
246         /*
247          * Release lock on inode
248          */
249         if (create)
250                 IWRITE_UNLOCK(ip);
251         else
252                 IREAD_UNLOCK(ip);
253         return rc;
254 }
255
256 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
257 {
258         return block_write_full_page(page, jfs_get_block, wbc);
259 }
260
261 static int jfs_writepages(struct address_space *mapping,
262                         struct writeback_control *wbc)
263 {
264         return mpage_writepages(mapping, wbc, jfs_get_block);
265 }
266
267 static int jfs_readpage(struct file *file, struct page *page)
268 {
269         return mpage_readpage(page, jfs_get_block);
270 }
271
272 static int jfs_readpages(struct file *file, struct address_space *mapping,
273                 struct list_head *pages, unsigned nr_pages)
274 {
275         return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
276 }
277
278 static int jfs_write_begin(struct file *file, struct address_space *mapping,
279                                 loff_t pos, unsigned len, unsigned flags,
280                                 struct page **pagep, void **fsdata)
281 {
282         return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
283                                 jfs_get_block);
284 }
285
286 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
287 {
288         return generic_block_bmap(mapping, block, jfs_get_block);
289 }
290
291 static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
292         const struct iovec *iov, loff_t offset, unsigned long nr_segs)
293 {
294         struct file *file = iocb->ki_filp;
295         struct inode *inode = file->f_mapping->host;
296
297         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
298                                 offset, nr_segs, jfs_get_block, NULL);
299 }
300
301 const struct address_space_operations jfs_aops = {
302         .readpage       = jfs_readpage,
303         .readpages      = jfs_readpages,
304         .writepage      = jfs_writepage,
305         .writepages     = jfs_writepages,
306         .sync_page      = block_sync_page,
307         .write_begin    = jfs_write_begin,
308         .write_end      = nobh_write_end,
309         .bmap           = jfs_bmap,
310         .direct_IO      = jfs_direct_IO,
311 };
312
313 /*
314  * Guts of jfs_truncate.  Called with locks already held.  Can be called
315  * with directory for truncating directory index table.
316  */
317 void jfs_truncate_nolock(struct inode *ip, loff_t length)
318 {
319         loff_t newsize;
320         tid_t tid;
321
322         ASSERT(length >= 0);
323
324         if (test_cflag(COMMIT_Nolink, ip)) {
325                 xtTruncate(0, ip, length, COMMIT_WMAP);
326                 return;
327         }
328
329         do {
330                 tid = txBegin(ip->i_sb, 0);
331
332                 /*
333                  * The commit_mutex cannot be taken before txBegin.
334                  * txBegin may block and there is a chance the inode
335                  * could be marked dirty and need to be committed
336                  * before txBegin unblocks
337                  */
338                 mutex_lock(&JFS_IP(ip)->commit_mutex);
339
340                 newsize = xtTruncate(tid, ip, length,
341                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
342                 if (newsize < 0) {
343                         txEnd(tid);
344                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
345                         break;
346                 }
347
348                 ip->i_mtime = ip->i_ctime = CURRENT_TIME;
349                 mark_inode_dirty(ip);
350
351                 txCommit(tid, 1, &ip, 0);
352                 txEnd(tid);
353                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
354         } while (newsize > length);     /* Truncate isn't always atomic */
355 }
356
357 void jfs_truncate(struct inode *ip)
358 {
359         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
360
361         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
362
363         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
364         jfs_truncate_nolock(ip, ip->i_size);
365         IWRITE_UNLOCK(ip);
366 }