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/uio.h>
26#include <linux/writeback.h>
27#include "jfs_incore.h"
28#include "jfs_inode.h"
29#include "jfs_filsys.h"
30#include "jfs_imap.h"
31#include "jfs_extent.h"
32#include "jfs_unicode.h"
33#include "jfs_debug.h"
34
35
36struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
37{
38	struct inode *inode;
39	int ret;
40
41	inode = iget_locked(sb, ino);
42	if (!inode)
43		return ERR_PTR(-ENOMEM);
44	if (!(inode->i_state & I_NEW))
45		return inode;
46
47	ret = diRead(inode);
48	if (ret < 0) {
49		iget_failed(inode);
50		return ERR_PTR(ret);
51	}
52
53	if (S_ISREG(inode->i_mode)) {
54		inode->i_op = &jfs_file_inode_operations;
55		inode->i_fop = &jfs_file_operations;
56		inode->i_mapping->a_ops = &jfs_aops;
57	} else if (S_ISDIR(inode->i_mode)) {
58		inode->i_op = &jfs_dir_inode_operations;
59		inode->i_fop = &jfs_dir_operations;
60	} else if (S_ISLNK(inode->i_mode)) {
61		if (inode->i_size >= IDATASIZE) {
62			inode->i_op = &page_symlink_inode_operations;
63			inode->i_mapping->a_ops = &jfs_aops;
64		} else {
65			inode->i_op = &jfs_fast_symlink_inode_operations;
66			/*
67			 * The inline data should be null-terminated, but
68			 * don't let on-disk corruption crash the kernel
69			 */
70			JFS_IP(inode)->i_inline[inode->i_size] = '\0';
71		}
72	} else {
73		inode->i_op = &jfs_file_inode_operations;
74		init_special_inode(inode, inode->i_mode, inode->i_rdev);
75	}
76	unlock_new_inode(inode);
77	return inode;
78}
79
80/*
81 * Workhorse of both fsync & write_inode
82 */
83int jfs_commit_inode(struct inode *inode, int wait)
84{
85	int rc = 0;
86	tid_t tid;
87	static int noisy = 5;
88
89	jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
90
91	/*
92	 * Don't commit if inode has been committed since last being
93	 * marked dirty, or if it has been deleted.
94	 */
95	if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
96		return 0;
97
98	if (isReadOnly(inode)) {
99		/* kernel allows writes to devices on read-only
100		 * partitions and may think inode is dirty
101		 */
102		if (!special_file(inode->i_mode) && noisy) {
103			jfs_err("jfs_commit_inode(0x%p) called on "
104				   "read-only volume", inode);
105			jfs_err("Is remount racy?");
106			noisy--;
107		}
108		return 0;
109	}
110
111	tid = txBegin(inode->i_sb, COMMIT_INODE);
112	mutex_lock(&JFS_IP(inode)->commit_mutex);
113
114	/*
115	 * Retest inode state after taking commit_mutex
116	 */
117	if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
118		rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
119
120	txEnd(tid);
121	mutex_unlock(&JFS_IP(inode)->commit_mutex);
122	return rc;
123}
124
125int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
126{
127	int wait = wbc->sync_mode == WB_SYNC_ALL;
128
129	if (inode->i_nlink == 0)
130		return 0;
131	/*
132	 * If COMMIT_DIRTY is not set, the inode isn't really dirty.
133	 * It has been committed since the last change, but was still
134	 * on the dirty inode list.
135	 */
136	 if (!test_cflag(COMMIT_Dirty, inode)) {
137		/* Make sure committed changes hit the disk */
138		jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
139		return 0;
140	 }
141
142	if (jfs_commit_inode(inode, wait)) {
143		jfs_err("jfs_write_inode: jfs_commit_inode failed!");
144		return -EIO;
145	} else
146		return 0;
147}
148
149void jfs_evict_inode(struct inode *inode)
150{
151	jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
152
153	if (!inode->i_nlink && !is_bad_inode(inode)) {
154		dquot_initialize(inode);
155
156		if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
157			truncate_inode_pages_final(&inode->i_data);
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		}
170	} else {
171		truncate_inode_pages_final(&inode->i_data);
172	}
173	clear_inode(inode);
174	dquot_drop(inode);
175}
176
177void jfs_dirty_inode(struct inode *inode, int flags)
178{
179	static int noisy = 5;
180
181	if (isReadOnly(inode)) {
182		if (!special_file(inode->i_mode) && noisy) {
183			/* kernel allows writes to devices on read-only
184			 * partitions and may try to mark inode dirty
185			 */
186			jfs_err("jfs_dirty_inode called on read-only volume");
187			jfs_err("Is remount racy?");
188			noisy--;
189		}
190		return;
191	}
192
193	set_cflag(COMMIT_Dirty, inode);
194}
195
196int jfs_get_block(struct inode *ip, sector_t lblock,
197		  struct buffer_head *bh_result, int create)
198{
199	s64 lblock64 = lblock;
200	int rc = 0;
201	xad_t xad;
202	s64 xaddr;
203	int xflag;
204	s32 xlen = bh_result->b_size >> ip->i_blkbits;
205
206	/*
207	 * Take appropriate lock on inode
208	 */
209	if (create)
210		IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
211	else
212		IREAD_LOCK(ip, RDWRLOCK_NORMAL);
213
214	if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
215	    (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
216	    xaddr) {
217		if (xflag & XAD_NOTRECORDED) {
218			if (!create)
219				/*
220				 * Allocated but not recorded, read treats
221				 * this as a hole
222				 */
223				goto unlock;
224#ifdef _JFS_4K
225			XADoffset(&xad, lblock64);
226			XADlength(&xad, xlen);
227			XADaddress(&xad, xaddr);
228#else				/* _JFS_4K */
229			/*
230			 * As long as block size = 4K, this isn't a problem.
231			 * We should mark the whole page not ABNR, but how
232			 * will we know to mark the other blocks BH_New?
233			 */
234			BUG();
235#endif				/* _JFS_4K */
236			rc = extRecord(ip, &xad);
237			if (rc)
238				goto unlock;
239			set_buffer_new(bh_result);
240		}
241
242		map_bh(bh_result, ip->i_sb, xaddr);
243		bh_result->b_size = xlen << ip->i_blkbits;
244		goto unlock;
245	}
246	if (!create)
247		goto unlock;
248
249	/*
250	 * Allocate a new block
251	 */
252#ifdef _JFS_4K
253	if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
254		goto unlock;
255	rc = extAlloc(ip, xlen, lblock64, &xad, false);
256	if (rc)
257		goto unlock;
258
259	set_buffer_new(bh_result);
260	map_bh(bh_result, ip->i_sb, addressXAD(&xad));
261	bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
262
263#else				/* _JFS_4K */
264	/*
265	 * We need to do whatever it takes to keep all but the last buffers
266	 * in 4K pages - see jfs_write.c
267	 */
268	BUG();
269#endif				/* _JFS_4K */
270
271      unlock:
272	/*
273	 * Release lock on inode
274	 */
275	if (create)
276		IWRITE_UNLOCK(ip);
277	else
278		IREAD_UNLOCK(ip);
279	return rc;
280}
281
282static int jfs_writepage(struct page *page, struct writeback_control *wbc)
283{
284	return block_write_full_page(page, jfs_get_block, wbc);
285}
286
287static int jfs_writepages(struct address_space *mapping,
288			struct writeback_control *wbc)
289{
290	return mpage_writepages(mapping, wbc, jfs_get_block);
291}
292
293static int jfs_readpage(struct file *file, struct page *page)
294{
295	return mpage_readpage(page, jfs_get_block);
296}
297
298static int jfs_readpages(struct file *file, struct address_space *mapping,
299		struct list_head *pages, unsigned nr_pages)
300{
301	return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
302}
303
304static void jfs_write_failed(struct address_space *mapping, loff_t to)
305{
306	struct inode *inode = mapping->host;
307
308	if (to > inode->i_size) {
309		truncate_pagecache(inode, inode->i_size);
310		jfs_truncate(inode);
311	}
312}
313
314static int jfs_write_begin(struct file *file, struct address_space *mapping,
315				loff_t pos, unsigned len, unsigned flags,
316				struct page **pagep, void **fsdata)
317{
318	int ret;
319
320	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
321				jfs_get_block);
322	if (unlikely(ret))
323		jfs_write_failed(mapping, pos + len);
324
325	return ret;
326}
327
328static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
329{
330	return generic_block_bmap(mapping, block, jfs_get_block);
331}
332
333static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
334			     loff_t offset)
335{
336	struct file *file = iocb->ki_filp;
337	struct address_space *mapping = file->f_mapping;
338	struct inode *inode = file->f_mapping->host;
339	size_t count = iov_iter_count(iter);
340	ssize_t ret;
341
342	ret = blockdev_direct_IO(iocb, inode, iter, offset, jfs_get_block);
343
344	/*
345	 * In case of error extending write may have instantiated a few
346	 * blocks outside i_size. Trim these off again.
347	 */
348	if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
349		loff_t isize = i_size_read(inode);
350		loff_t end = offset + count;
351
352		if (end > isize)
353			jfs_write_failed(mapping, end);
354	}
355
356	return ret;
357}
358
359const struct address_space_operations jfs_aops = {
360	.readpage	= jfs_readpage,
361	.readpages	= jfs_readpages,
362	.writepage	= jfs_writepage,
363	.writepages	= jfs_writepages,
364	.write_begin	= jfs_write_begin,
365	.write_end	= nobh_write_end,
366	.bmap		= jfs_bmap,
367	.direct_IO	= jfs_direct_IO,
368};
369
370/*
371 * Guts of jfs_truncate.  Called with locks already held.  Can be called
372 * with directory for truncating directory index table.
373 */
374void jfs_truncate_nolock(struct inode *ip, loff_t length)
375{
376	loff_t newsize;
377	tid_t tid;
378
379	ASSERT(length >= 0);
380
381	if (test_cflag(COMMIT_Nolink, ip)) {
382		xtTruncate(0, ip, length, COMMIT_WMAP);
383		return;
384	}
385
386	do {
387		tid = txBegin(ip->i_sb, 0);
388
389		/*
390		 * The commit_mutex cannot be taken before txBegin.
391		 * txBegin may block and there is a chance the inode
392		 * could be marked dirty and need to be committed
393		 * before txBegin unblocks
394		 */
395		mutex_lock(&JFS_IP(ip)->commit_mutex);
396
397		newsize = xtTruncate(tid, ip, length,
398				     COMMIT_TRUNCATE | COMMIT_PWMAP);
399		if (newsize < 0) {
400			txEnd(tid);
401			mutex_unlock(&JFS_IP(ip)->commit_mutex);
402			break;
403		}
404
405		ip->i_mtime = ip->i_ctime = CURRENT_TIME;
406		mark_inode_dirty(ip);
407
408		txCommit(tid, 1, &ip, 0);
409		txEnd(tid);
410		mutex_unlock(&JFS_IP(ip)->commit_mutex);
411	} while (newsize > length);	/* Truncate isn't always atomic */
412}
413
414void jfs_truncate(struct inode *ip)
415{
416	jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
417
418	nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
419
420	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
421	jfs_truncate_nolock(ip, ip->i_size);
422	IWRITE_UNLOCK(ip);
423}
424