xref: /openbmc/linux/fs/jfs/inode.c (revision 92b19ff5) 
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 
36 struct 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 			inode->i_link = JFS_IP(inode)->i_inline;
67 			/*
68 			 * The inline data should be null-terminated, but
69 			 * don't let on-disk corruption crash the kernel
70 			 */
71 			inode->i_link[inode->i_size] = '\0';
72 		}
73 	} else {
74 		inode->i_op = &jfs_file_inode_operations;
75 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
76 	}
77 	unlock_new_inode(inode);
78 	return inode;
79 }
80 
81 /*
82  * Workhorse of both fsync & write_inode
83  */
84 int jfs_commit_inode(struct inode *inode, int wait)
85 {
86 	int rc = 0;
87 	tid_t tid;
88 	static int noisy = 5;
89 
90 	jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
91 
92 	/*
93 	 * Don't commit if inode has been committed since last being
94 	 * marked dirty, or if it has been deleted.
95 	 */
96 	if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
97 		return 0;
98 
99 	if (isReadOnly(inode)) {
100 		/* kernel allows writes to devices on read-only
101 		 * partitions and may think inode is dirty
102 		 */
103 		if (!special_file(inode->i_mode) && noisy) {
104 			jfs_err("jfs_commit_inode(0x%p) called on "
105 				   "read-only volume", inode);
106 			jfs_err("Is remount racy?");
107 			noisy--;
108 		}
109 		return 0;
110 	}
111 
112 	tid = txBegin(inode->i_sb, COMMIT_INODE);
113 	mutex_lock(&JFS_IP(inode)->commit_mutex);
114 
115 	/*
116 	 * Retest inode state after taking commit_mutex
117 	 */
118 	if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
119 		rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
120 
121 	txEnd(tid);
122 	mutex_unlock(&JFS_IP(inode)->commit_mutex);
123 	return rc;
124 }
125 
126 int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
127 {
128 	int wait = wbc->sync_mode == WB_SYNC_ALL;
129 
130 	if (inode->i_nlink == 0)
131 		return 0;
132 	/*
133 	 * If COMMIT_DIRTY is not set, the inode isn't really dirty.
134 	 * It has been committed since the last change, but was still
135 	 * on the dirty inode list.
136 	 */
137 	if (!test_cflag(COMMIT_Dirty, inode)) {
138 		/* Make sure committed changes hit the disk */
139 		jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
140 		return 0;
141 	}
142 
143 	if (jfs_commit_inode(inode, wait)) {
144 		jfs_err("jfs_write_inode: jfs_commit_inode failed!");
145 		return -EIO;
146 	} else
147 		return 0;
148 }
149 
150 void jfs_evict_inode(struct inode *inode)
151 {
152 	jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
153 
154 	if (!inode->i_nlink && !is_bad_inode(inode)) {
155 		dquot_initialize(inode);
156 
157 		if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
158 			truncate_inode_pages_final(&inode->i_data);
159 
160 			if (test_cflag(COMMIT_Freewmap, inode))
161 				jfs_free_zero_link(inode);
162 
163 			diFree(inode);
164 
165 			/*
166 			 * Free the inode from the quota allocation.
167 			 */
168 			dquot_initialize(inode);
169 			dquot_free_inode(inode);
170 		}
171 	} else {
172 		truncate_inode_pages_final(&inode->i_data);
173 	}
174 	clear_inode(inode);
175 	dquot_drop(inode);
176 }
177 
178 void jfs_dirty_inode(struct inode *inode, int flags)
179 {
180 	static int noisy = 5;
181 
182 	if (isReadOnly(inode)) {
183 		if (!special_file(inode->i_mode) && noisy) {
184 			/* kernel allows writes to devices on read-only
185 			 * partitions and may try to mark inode dirty
186 			 */
187 			jfs_err("jfs_dirty_inode called on read-only volume");
188 			jfs_err("Is remount racy?");
189 			noisy--;
190 		}
191 		return;
192 	}
193 
194 	set_cflag(COMMIT_Dirty, inode);
195 }
196 
197 int jfs_get_block(struct inode *ip, sector_t lblock,
198 		  struct buffer_head *bh_result, int create)
199 {
200 	s64 lblock64 = lblock;
201 	int rc = 0;
202 	xad_t xad;
203 	s64 xaddr;
204 	int xflag;
205 	s32 xlen = bh_result->b_size >> ip->i_blkbits;
206 
207 	/*
208 	 * Take appropriate lock on inode
209 	 */
210 	if (create)
211 		IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
212 	else
213 		IREAD_LOCK(ip, RDWRLOCK_NORMAL);
214 
215 	if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
216 	    (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
217 	    xaddr) {
218 		if (xflag & XAD_NOTRECORDED) {
219 			if (!create)
220 				/*
221 				 * Allocated but not recorded, read treats
222 				 * this as a hole
223 				 */
224 				goto unlock;
225 #ifdef _JFS_4K
226 			XADoffset(&xad, lblock64);
227 			XADlength(&xad, xlen);
228 			XADaddress(&xad, xaddr);
229 #else				/* _JFS_4K */
230 			/*
231 			 * As long as block size = 4K, this isn't a problem.
232 			 * We should mark the whole page not ABNR, but how
233 			 * will we know to mark the other blocks BH_New?
234 			 */
235 			BUG();
236 #endif				/* _JFS_4K */
237 			rc = extRecord(ip, &xad);
238 			if (rc)
239 				goto unlock;
240 			set_buffer_new(bh_result);
241 		}
242 
243 		map_bh(bh_result, ip->i_sb, xaddr);
244 		bh_result->b_size = xlen << ip->i_blkbits;
245 		goto unlock;
246 	}
247 	if (!create)
248 		goto unlock;
249 
250 	/*
251 	 * Allocate a new block
252 	 */
253 #ifdef _JFS_4K
254 	if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
255 		goto unlock;
256 	rc = extAlloc(ip, xlen, lblock64, &xad, false);
257 	if (rc)
258 		goto unlock;
259 
260 	set_buffer_new(bh_result);
261 	map_bh(bh_result, ip->i_sb, addressXAD(&xad));
262 	bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
263 
264 #else				/* _JFS_4K */
265 	/*
266 	 * We need to do whatever it takes to keep all but the last buffers
267 	 * in 4K pages - see jfs_write.c
268 	 */
269 	BUG();
270 #endif				/* _JFS_4K */
271 
272       unlock:
273 	/*
274 	 * Release lock on inode
275 	 */
276 	if (create)
277 		IWRITE_UNLOCK(ip);
278 	else
279 		IREAD_UNLOCK(ip);
280 	return rc;
281 }
282 
283 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
284 {
285 	return block_write_full_page(page, jfs_get_block, wbc);
286 }
287 
288 static int jfs_writepages(struct address_space *mapping,
289 			struct writeback_control *wbc)
290 {
291 	return mpage_writepages(mapping, wbc, jfs_get_block);
292 }
293 
294 static int jfs_readpage(struct file *file, struct page *page)
295 {
296 	return mpage_readpage(page, jfs_get_block);
297 }
298 
299 static int jfs_readpages(struct file *file, struct address_space *mapping,
300 		struct list_head *pages, unsigned nr_pages)
301 {
302 	return mpage_readpages(mapping, pages, nr_pages, jfs_get_block);
303 }
304 
305 static void jfs_write_failed(struct address_space *mapping, loff_t to)
306 {
307 	struct inode *inode = mapping->host;
308 
309 	if (to > inode->i_size) {
310 		truncate_pagecache(inode, inode->i_size);
311 		jfs_truncate(inode);
312 	}
313 }
314 
315 static int jfs_write_begin(struct file *file, struct address_space *mapping,
316 				loff_t pos, unsigned len, unsigned flags,
317 				struct page **pagep, void **fsdata)
318 {
319 	int ret;
320 
321 	ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
322 				jfs_get_block);
323 	if (unlikely(ret))
324 		jfs_write_failed(mapping, pos + len);
325 
326 	return ret;
327 }
328 
329 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
330 {
331 	return generic_block_bmap(mapping, block, jfs_get_block);
332 }
333 
334 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
335 			     loff_t offset)
336 {
337 	struct file *file = iocb->ki_filp;
338 	struct address_space *mapping = file->f_mapping;
339 	struct inode *inode = file->f_mapping->host;
340 	size_t count = iov_iter_count(iter);
341 	ssize_t ret;
342 
343 	ret = blockdev_direct_IO(iocb, inode, iter, offset, jfs_get_block);
344 
345 	/*
346 	 * In case of error extending write may have instantiated a few
347 	 * blocks outside i_size. Trim these off again.
348 	 */
349 	if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
350 		loff_t isize = i_size_read(inode);
351 		loff_t end = offset + count;
352 
353 		if (end > isize)
354 			jfs_write_failed(mapping, end);
355 	}
356 
357 	return ret;
358 }
359 
360 const struct address_space_operations jfs_aops = {
361 	.readpage	= jfs_readpage,
362 	.readpages	= jfs_readpages,
363 	.writepage	= jfs_writepage,
364 	.writepages	= jfs_writepages,
365 	.write_begin	= jfs_write_begin,
366 	.write_end	= nobh_write_end,
367 	.bmap		= jfs_bmap,
368 	.direct_IO	= jfs_direct_IO,
369 };
370 
371 /*
372  * Guts of jfs_truncate.  Called with locks already held.  Can be called
373  * with directory for truncating directory index table.
374  */
375 void jfs_truncate_nolock(struct inode *ip, loff_t length)
376 {
377 	loff_t newsize;
378 	tid_t tid;
379 
380 	ASSERT(length >= 0);
381 
382 	if (test_cflag(COMMIT_Nolink, ip)) {
383 		xtTruncate(0, ip, length, COMMIT_WMAP);
384 		return;
385 	}
386 
387 	do {
388 		tid = txBegin(ip->i_sb, 0);
389 
390 		/*
391 		 * The commit_mutex cannot be taken before txBegin.
392 		 * txBegin may block and there is a chance the inode
393 		 * could be marked dirty and need to be committed
394 		 * before txBegin unblocks
395 		 */
396 		mutex_lock(&JFS_IP(ip)->commit_mutex);
397 
398 		newsize = xtTruncate(tid, ip, length,
399 				     COMMIT_TRUNCATE | COMMIT_PWMAP);
400 		if (newsize < 0) {
401 			txEnd(tid);
402 			mutex_unlock(&JFS_IP(ip)->commit_mutex);
403 			break;
404 		}
405 
406 		ip->i_mtime = ip->i_ctime = CURRENT_TIME;
407 		mark_inode_dirty(ip);
408 
409 		txCommit(tid, 1, &ip, 0);
410 		txEnd(tid);
411 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
412 	} while (newsize > length);	/* Truncate isn't always atomic */
413 }
414 
415 void jfs_truncate(struct inode *ip)
416 {
417 	jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
418 
419 	nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
420 
421 	IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
422 	jfs_truncate_nolock(ip, ip->i_size);
423 	IWRITE_UNLOCK(ip);
424 }
425