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
35struct 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_fast_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 */
82int 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
124int 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
148void jfs_evict_inode(struct inode *inode)
149{
150	jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
151
152	if (!inode->i_nlink && !is_bad_inode(inode)) {
153		dquot_initialize(inode);
154
155		if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
156			truncate_inode_pages(&inode->i_data, 0);
157
158			if (test_cflag(COMMIT_Freewmap, inode))
159				jfs_free_zero_link(inode);
160
161			diFree(inode);
162
163			/*
164			 * Free the inode from the quota allocation.
165			 */
166			dquot_initialize(inode);
167			dquot_free_inode(inode);
168		}
169	} else {
170		truncate_inode_pages(&inode->i_data, 0);
171	}
172	end_writeback(inode);
173	dquot_drop(inode);
174}
175
176void jfs_dirty_inode(struct inode *inode)
177{
178	static int noisy = 5;
179
180	if (isReadOnly(inode)) {
181		if (!special_file(inode->i_mode) && noisy) {
182			/* kernel allows writes to devices on read-only
183			 * partitions and may try to mark inode dirty
184			 */
185			jfs_err("jfs_dirty_inode called on read-only volume");
186			jfs_err("Is remount racy?");
187			noisy--;
188		}
189		return;
190	}
191
192	set_cflag(COMMIT_Dirty, inode);
193}
194
195int jfs_get_block(struct inode *ip, sector_t lblock,
196		  struct buffer_head *bh_result, int create)
197{
198	s64 lblock64 = lblock;
199	int rc = 0;
200	xad_t xad;
201	s64 xaddr;
202	int xflag;
203	s32 xlen = bh_result->b_size >> ip->i_blkbits;
204
205	/*
206	 * Take appropriate lock on inode
207	 */
208	if (create)
209		IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
210	else
211		IREAD_LOCK(ip, RDWRLOCK_NORMAL);
212
213	if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
214	    (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
215	    xaddr) {
216		if (xflag & XAD_NOTRECORDED) {
217			if (!create)
218				/*
219				 * Allocated but not recorded, read treats
220				 * this as a hole
221				 */
222				goto unlock;
223#ifdef _JFS_4K
224			XADoffset(&xad, lblock64);
225			XADlength(&xad, xlen);
226			XADaddress(&xad, xaddr);
227#else				/* _JFS_4K */
228			/*
229			 * As long as block size = 4K, this isn't a problem.
230			 * We should mark the whole page not ABNR, but how
231			 * will we know to mark the other blocks BH_New?
232			 */
233			BUG();
234#endif				/* _JFS_4K */
235			rc = extRecord(ip, &xad);
236			if (rc)
237				goto unlock;
238			set_buffer_new(bh_result);
239		}
240
241		map_bh(bh_result, ip->i_sb, xaddr);
242		bh_result->b_size = xlen << ip->i_blkbits;
243		goto unlock;
244	}
245	if (!create)
246		goto unlock;
247
248	/*
249	 * Allocate a new block
250	 */
251#ifdef _JFS_4K
252	if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
253		goto unlock;
254	rc = extAlloc(ip, xlen, lblock64, &xad, false);
255	if (rc)
256		goto unlock;
257
258	set_buffer_new(bh_result);
259	map_bh(bh_result, ip->i_sb, addressXAD(&xad));
260	bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
261
262#else				/* _JFS_4K */
263	/*
264	 * We need to do whatever it takes to keep all but the last buffers
265	 * in 4K pages - see jfs_write.c
266	 */
267	BUG();
268#endif				/* _JFS_4K */
269
270      unlock:
271	/*
272	 * Release lock on inode
273	 */
274	if (create)
275		IWRITE_UNLOCK(ip);
276	else
277		IREAD_UNLOCK(ip);
278	return rc;
279}
280
281static int jfs_writepage(struct page *page, struct writeback_control *wbc)
282{
283	return block_write_full_page(page, jfs_get_block, wbc);
284}
285
286static int jfs_writepages(struct address_space *mapping,
287			struct writeback_control *wbc)
288{
289	return mpage_writepages(mapping, wbc, jfs_get_block);
290}
291
292static int jfs_readpage(struct file *file, struct page *page)
293{
294	return mpage_readpage(page, jfs_get_block);
295}
296
297static int jfs_readpages(struct file *file, struct address_space *mapping,
298		struct list_head *pages, unsigned nr_pages)
299{
300	return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
301}
302
303static int jfs_write_begin(struct file *file, struct address_space *mapping,
304				loff_t pos, unsigned len, unsigned flags,
305				struct page **pagep, void **fsdata)
306{
307	int ret;
308
309	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
310				jfs_get_block);
311	if (unlikely(ret)) {
312		loff_t isize = mapping->host->i_size;
313		if (pos + len > isize)
314			vmtruncate(mapping->host, isize);
315	}
316
317	return ret;
318}
319
320static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
321{
322	return generic_block_bmap(mapping, block, jfs_get_block);
323}
324
325static ssize_t jfs_direct_IO(int rw, struct kiocb *iocb,
326	const struct iovec *iov, loff_t offset, unsigned long nr_segs)
327{
328	struct file *file = iocb->ki_filp;
329	struct inode *inode = file->f_mapping->host;
330	ssize_t ret;
331
332	ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
333				offset, nr_segs, jfs_get_block, NULL);
334
335	/*
336	 * In case of error extending write may have instantiated a few
337	 * blocks outside i_size. Trim these off again.
338	 */
339	if (unlikely((rw & WRITE) && ret < 0)) {
340		loff_t isize = i_size_read(inode);
341		loff_t end = offset + iov_length(iov, nr_segs);
342
343		if (end > isize)
344			vmtruncate(inode, isize);
345	}
346
347	return ret;
348}
349
350const struct address_space_operations jfs_aops = {
351	.readpage	= jfs_readpage,
352	.readpages	= jfs_readpages,
353	.writepage	= jfs_writepage,
354	.writepages	= jfs_writepages,
355	.sync_page	= block_sync_page,
356	.write_begin	= jfs_write_begin,
357	.write_end	= nobh_write_end,
358	.bmap		= jfs_bmap,
359	.direct_IO	= jfs_direct_IO,
360};
361
362/*
363 * Guts of jfs_truncate.  Called with locks already held.  Can be called
364 * with directory for truncating directory index table.
365 */
366void jfs_truncate_nolock(struct inode *ip, loff_t length)
367{
368	loff_t newsize;
369	tid_t tid;
370
371	ASSERT(length >= 0);
372
373	if (test_cflag(COMMIT_Nolink, ip)) {
374		xtTruncate(0, ip, length, COMMIT_WMAP);
375		return;
376	}
377
378	do {
379		tid = txBegin(ip->i_sb, 0);
380
381		/*
382		 * The commit_mutex cannot be taken before txBegin.
383		 * txBegin may block and there is a chance the inode
384		 * could be marked dirty and need to be committed
385		 * before txBegin unblocks
386		 */
387		mutex_lock(&JFS_IP(ip)->commit_mutex);
388
389		newsize = xtTruncate(tid, ip, length,
390				     COMMIT_TRUNCATE | COMMIT_PWMAP);
391		if (newsize < 0) {
392			txEnd(tid);
393			mutex_unlock(&JFS_IP(ip)->commit_mutex);
394			break;
395		}
396
397		ip->i_mtime = ip->i_ctime = CURRENT_TIME;
398		mark_inode_dirty(ip);
399
400		txCommit(tid, 1, &ip, 0);
401		txEnd(tid);
402		mutex_unlock(&JFS_IP(ip)->commit_mutex);
403	} while (newsize > length);	/* Truncate isn't always atomic */
404}
405
406void jfs_truncate(struct inode *ip)
407{
408	jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
409
410	nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
411
412	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
413	jfs_truncate_nolock(ip, ip->i_size);
414	IWRITE_UNLOCK(ip);
415}
416