xref: /openbmc/linux/fs/nilfs2/inode.c (revision 155130a4f7848b1aac439cab6bda1a175507c71c)
1 /*
2  * inode.c - NILFS inode operations.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  *
22  */
23 
24 #include <linux/buffer_head.h>
25 #include <linux/gfp.h>
26 #include <linux/mpage.h>
27 #include <linux/writeback.h>
28 #include <linux/uio.h>
29 #include "nilfs.h"
30 #include "segment.h"
31 #include "page.h"
32 #include "mdt.h"
33 #include "cpfile.h"
34 #include "ifile.h"
35 
36 
37 /**
38  * nilfs_get_block() - get a file block on the filesystem (callback function)
39  * @inode - inode struct of the target file
40  * @blkoff - file block number
41  * @bh_result - buffer head to be mapped on
42  * @create - indicate whether allocating the block or not when it has not
43  *      been allocated yet.
44  *
45  * This function does not issue actual read request of the specified data
46  * block. It is done by VFS.
47  */
48 int nilfs_get_block(struct inode *inode, sector_t blkoff,
49 		    struct buffer_head *bh_result, int create)
50 {
51 	struct nilfs_inode_info *ii = NILFS_I(inode);
52 	__u64 blknum = 0;
53 	int err = 0, ret;
54 	struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
55 	unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
56 
57 	down_read(&NILFS_MDT(dat)->mi_sem);
58 	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
59 	up_read(&NILFS_MDT(dat)->mi_sem);
60 	if (ret >= 0) {	/* found */
61 		map_bh(bh_result, inode->i_sb, blknum);
62 		if (ret > 0)
63 			bh_result->b_size = (ret << inode->i_blkbits);
64 		goto out;
65 	}
66 	/* data block was not found */
67 	if (ret == -ENOENT && create) {
68 		struct nilfs_transaction_info ti;
69 
70 		bh_result->b_blocknr = 0;
71 		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
72 		if (unlikely(err))
73 			goto out;
74 		err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
75 					(unsigned long)bh_result);
76 		if (unlikely(err != 0)) {
77 			if (err == -EEXIST) {
78 				/*
79 				 * The get_block() function could be called
80 				 * from multiple callers for an inode.
81 				 * However, the page having this block must
82 				 * be locked in this case.
83 				 */
84 				printk(KERN_WARNING
85 				       "nilfs_get_block: a race condition "
86 				       "while inserting a data block. "
87 				       "(inode number=%lu, file block "
88 				       "offset=%llu)\n",
89 				       inode->i_ino,
90 				       (unsigned long long)blkoff);
91 				err = 0;
92 			} else if (err == -EINVAL) {
93 				nilfs_error(inode->i_sb, __func__,
94 					    "broken bmap (inode=%lu)\n",
95 					    inode->i_ino);
96 				err = -EIO;
97 			}
98 			nilfs_transaction_abort(inode->i_sb);
99 			goto out;
100 		}
101 		nilfs_mark_inode_dirty(inode);
102 		nilfs_transaction_commit(inode->i_sb); /* never fails */
103 		/* Error handling should be detailed */
104 		set_buffer_new(bh_result);
105 		map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
106 						      to proper value */
107 	} else if (ret == -ENOENT) {
108 		/* not found is not error (e.g. hole); must return without
109 		   the mapped state flag. */
110 		;
111 	} else {
112 		err = ret;
113 	}
114 
115  out:
116 	return err;
117 }
118 
119 /**
120  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
121  * address_space_operations.
122  * @file - file struct of the file to be read
123  * @page - the page to be read
124  */
125 static int nilfs_readpage(struct file *file, struct page *page)
126 {
127 	return mpage_readpage(page, nilfs_get_block);
128 }
129 
130 /**
131  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
132  * address_space_operations.
133  * @file - file struct of the file to be read
134  * @mapping - address_space struct used for reading multiple pages
135  * @pages - the pages to be read
136  * @nr_pages - number of pages to be read
137  */
138 static int nilfs_readpages(struct file *file, struct address_space *mapping,
139 			   struct list_head *pages, unsigned nr_pages)
140 {
141 	return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
142 }
143 
144 static int nilfs_writepages(struct address_space *mapping,
145 			    struct writeback_control *wbc)
146 {
147 	struct inode *inode = mapping->host;
148 	int err = 0;
149 
150 	if (wbc->sync_mode == WB_SYNC_ALL)
151 		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
152 						    wbc->range_start,
153 						    wbc->range_end);
154 	return err;
155 }
156 
157 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
158 {
159 	struct inode *inode = page->mapping->host;
160 	int err;
161 
162 	redirty_page_for_writepage(wbc, page);
163 	unlock_page(page);
164 
165 	if (wbc->sync_mode == WB_SYNC_ALL) {
166 		err = nilfs_construct_segment(inode->i_sb);
167 		if (unlikely(err))
168 			return err;
169 	} else if (wbc->for_reclaim)
170 		nilfs_flush_segment(inode->i_sb, inode->i_ino);
171 
172 	return 0;
173 }
174 
175 static int nilfs_set_page_dirty(struct page *page)
176 {
177 	int ret = __set_page_dirty_buffers(page);
178 
179 	if (ret) {
180 		struct inode *inode = page->mapping->host;
181 		struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
182 		unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
183 
184 		nilfs_set_file_dirty(sbi, inode, nr_dirty);
185 	}
186 	return ret;
187 }
188 
189 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
190 			     loff_t pos, unsigned len, unsigned flags,
191 			     struct page **pagep, void **fsdata)
192 
193 {
194 	struct inode *inode = mapping->host;
195 	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
196 
197 	if (unlikely(err))
198 		return err;
199 
200 	err = block_write_begin(mapping, pos, len, flags, pagep,
201 				nilfs_get_block);
202 	if (unlikely(err)) {
203 		loff_t isize = mapping->host->i_size;
204 		if (pos + len > isize)
205 			vmtruncate(mapping->host, isize);
206 
207 		nilfs_transaction_abort(inode->i_sb);
208 	}
209 	return err;
210 }
211 
212 static int nilfs_write_end(struct file *file, struct address_space *mapping,
213 			   loff_t pos, unsigned len, unsigned copied,
214 			   struct page *page, void *fsdata)
215 {
216 	struct inode *inode = mapping->host;
217 	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
218 	unsigned nr_dirty;
219 	int err;
220 
221 	nr_dirty = nilfs_page_count_clean_buffers(page, start,
222 						  start + copied);
223 	copied = generic_write_end(file, mapping, pos, len, copied, page,
224 				   fsdata);
225 	nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
226 	err = nilfs_transaction_commit(inode->i_sb);
227 	return err ? : copied;
228 }
229 
230 static ssize_t
231 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
232 		loff_t offset, unsigned long nr_segs)
233 {
234 	struct file *file = iocb->ki_filp;
235 	struct inode *inode = file->f_mapping->host;
236 	ssize_t size;
237 
238 	if (rw == WRITE)
239 		return 0;
240 
241 	/* Needs synchronization with the cleaner */
242 	size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
243 				  offset, nr_segs, nilfs_get_block, NULL);
244 
245 	/*
246 	 * In case of error extending write may have instantiated a few
247 	 * blocks outside i_size. Trim these off again.
248 	 */
249 	if (unlikely((rw & WRITE) && size < 0)) {
250 		loff_t isize = i_size_read(inode);
251 		loff_t end = offset + iov_length(iov, nr_segs);
252 
253 		if (end > isize)
254 			vmtruncate(inode, isize);
255 	}
256 
257 	return size;
258 }
259 
260 const struct address_space_operations nilfs_aops = {
261 	.writepage		= nilfs_writepage,
262 	.readpage		= nilfs_readpage,
263 	.sync_page		= block_sync_page,
264 	.writepages		= nilfs_writepages,
265 	.set_page_dirty		= nilfs_set_page_dirty,
266 	.readpages		= nilfs_readpages,
267 	.write_begin		= nilfs_write_begin,
268 	.write_end		= nilfs_write_end,
269 	/* .releasepage		= nilfs_releasepage, */
270 	.invalidatepage		= block_invalidatepage,
271 	.direct_IO		= nilfs_direct_IO,
272 	.is_partially_uptodate  = block_is_partially_uptodate,
273 };
274 
275 struct inode *nilfs_new_inode(struct inode *dir, int mode)
276 {
277 	struct super_block *sb = dir->i_sb;
278 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
279 	struct inode *inode;
280 	struct nilfs_inode_info *ii;
281 	int err = -ENOMEM;
282 	ino_t ino;
283 
284 	inode = new_inode(sb);
285 	if (unlikely(!inode))
286 		goto failed;
287 
288 	mapping_set_gfp_mask(inode->i_mapping,
289 			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
290 
291 	ii = NILFS_I(inode);
292 	ii->i_state = 1 << NILFS_I_NEW;
293 
294 	err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
295 	if (unlikely(err))
296 		goto failed_ifile_create_inode;
297 	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
298 
299 	atomic_inc(&sbi->s_inodes_count);
300 	inode_init_owner(inode, dir, mode);
301 	inode->i_ino = ino;
302 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
303 
304 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
305 		err = nilfs_bmap_read(ii->i_bmap, NULL);
306 		if (err < 0)
307 			goto failed_bmap;
308 
309 		set_bit(NILFS_I_BMAP, &ii->i_state);
310 		/* No lock is needed; iget() ensures it. */
311 	}
312 
313 	ii->i_flags = NILFS_I(dir)->i_flags;
314 	if (S_ISLNK(mode))
315 		ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
316 	if (!S_ISDIR(mode))
317 		ii->i_flags &= ~NILFS_DIRSYNC_FL;
318 
319 	/* ii->i_file_acl = 0; */
320 	/* ii->i_dir_acl = 0; */
321 	ii->i_dir_start_lookup = 0;
322 	ii->i_cno = 0;
323 	nilfs_set_inode_flags(inode);
324 	spin_lock(&sbi->s_next_gen_lock);
325 	inode->i_generation = sbi->s_next_generation++;
326 	spin_unlock(&sbi->s_next_gen_lock);
327 	insert_inode_hash(inode);
328 
329 	err = nilfs_init_acl(inode, dir);
330 	if (unlikely(err))
331 		goto failed_acl; /* never occur. When supporting
332 				    nilfs_init_acl(), proper cancellation of
333 				    above jobs should be considered */
334 
335 	return inode;
336 
337  failed_acl:
338  failed_bmap:
339 	inode->i_nlink = 0;
340 	iput(inode);  /* raw_inode will be deleted through
341 			 generic_delete_inode() */
342 	goto failed;
343 
344  failed_ifile_create_inode:
345 	make_bad_inode(inode);
346 	iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
347 			 called */
348  failed:
349 	return ERR_PTR(err);
350 }
351 
352 void nilfs_free_inode(struct inode *inode)
353 {
354 	struct super_block *sb = inode->i_sb;
355 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
356 
357 	clear_inode(inode);
358 	/* XXX: check error code? Is there any thing I can do? */
359 	(void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
360 	atomic_dec(&sbi->s_inodes_count);
361 }
362 
363 void nilfs_set_inode_flags(struct inode *inode)
364 {
365 	unsigned int flags = NILFS_I(inode)->i_flags;
366 
367 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
368 			    S_DIRSYNC);
369 	if (flags & NILFS_SYNC_FL)
370 		inode->i_flags |= S_SYNC;
371 	if (flags & NILFS_APPEND_FL)
372 		inode->i_flags |= S_APPEND;
373 	if (flags & NILFS_IMMUTABLE_FL)
374 		inode->i_flags |= S_IMMUTABLE;
375 #ifndef NILFS_ATIME_DISABLE
376 	if (flags & NILFS_NOATIME_FL)
377 #endif
378 		inode->i_flags |= S_NOATIME;
379 	if (flags & NILFS_DIRSYNC_FL)
380 		inode->i_flags |= S_DIRSYNC;
381 	mapping_set_gfp_mask(inode->i_mapping,
382 			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
383 }
384 
385 int nilfs_read_inode_common(struct inode *inode,
386 			    struct nilfs_inode *raw_inode)
387 {
388 	struct nilfs_inode_info *ii = NILFS_I(inode);
389 	int err;
390 
391 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
392 	inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
393 	inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
394 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
395 	inode->i_size = le64_to_cpu(raw_inode->i_size);
396 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
397 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
398 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
399 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
400 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
401 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
402 	if (inode->i_nlink == 0 && inode->i_mode == 0)
403 		return -EINVAL; /* this inode is deleted */
404 
405 	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
406 	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
407 #if 0
408 	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
409 	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
410 		0 : le32_to_cpu(raw_inode->i_dir_acl);
411 #endif
412 	ii->i_dir_start_lookup = 0;
413 	ii->i_cno = 0;
414 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
415 
416 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
417 	    S_ISLNK(inode->i_mode)) {
418 		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
419 		if (err < 0)
420 			return err;
421 		set_bit(NILFS_I_BMAP, &ii->i_state);
422 		/* No lock is needed; iget() ensures it. */
423 	}
424 	return 0;
425 }
426 
427 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
428 			      struct inode *inode)
429 {
430 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
431 	struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
432 	struct buffer_head *bh;
433 	struct nilfs_inode *raw_inode;
434 	int err;
435 
436 	down_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
437 	err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
438 	if (unlikely(err))
439 		goto bad_inode;
440 
441 	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
442 
443 	err = nilfs_read_inode_common(inode, raw_inode);
444 	if (err)
445 		goto failed_unmap;
446 
447 	if (S_ISREG(inode->i_mode)) {
448 		inode->i_op = &nilfs_file_inode_operations;
449 		inode->i_fop = &nilfs_file_operations;
450 		inode->i_mapping->a_ops = &nilfs_aops;
451 	} else if (S_ISDIR(inode->i_mode)) {
452 		inode->i_op = &nilfs_dir_inode_operations;
453 		inode->i_fop = &nilfs_dir_operations;
454 		inode->i_mapping->a_ops = &nilfs_aops;
455 	} else if (S_ISLNK(inode->i_mode)) {
456 		inode->i_op = &nilfs_symlink_inode_operations;
457 		inode->i_mapping->a_ops = &nilfs_aops;
458 	} else {
459 		inode->i_op = &nilfs_special_inode_operations;
460 		init_special_inode(
461 			inode, inode->i_mode,
462 			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
463 	}
464 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
465 	brelse(bh);
466 	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
467 	nilfs_set_inode_flags(inode);
468 	return 0;
469 
470  failed_unmap:
471 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
472 	brelse(bh);
473 
474  bad_inode:
475 	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
476 	return err;
477 }
478 
479 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
480 {
481 	struct inode *inode;
482 	int err;
483 
484 	inode = iget_locked(sb, ino);
485 	if (unlikely(!inode))
486 		return ERR_PTR(-ENOMEM);
487 	if (!(inode->i_state & I_NEW))
488 		return inode;
489 
490 	err = __nilfs_read_inode(sb, ino, inode);
491 	if (unlikely(err)) {
492 		iget_failed(inode);
493 		return ERR_PTR(err);
494 	}
495 	unlock_new_inode(inode);
496 	return inode;
497 }
498 
499 void nilfs_write_inode_common(struct inode *inode,
500 			      struct nilfs_inode *raw_inode, int has_bmap)
501 {
502 	struct nilfs_inode_info *ii = NILFS_I(inode);
503 
504 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
505 	raw_inode->i_uid = cpu_to_le32(inode->i_uid);
506 	raw_inode->i_gid = cpu_to_le32(inode->i_gid);
507 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
508 	raw_inode->i_size = cpu_to_le64(inode->i_size);
509 	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
510 	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
511 	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
512 	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
513 	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
514 
515 	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
516 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
517 
518 	if (has_bmap)
519 		nilfs_bmap_write(ii->i_bmap, raw_inode);
520 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
521 		raw_inode->i_device_code =
522 			cpu_to_le64(huge_encode_dev(inode->i_rdev));
523 	/* When extending inode, nilfs->ns_inode_size should be checked
524 	   for substitutions of appended fields */
525 }
526 
527 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
528 {
529 	ino_t ino = inode->i_ino;
530 	struct nilfs_inode_info *ii = NILFS_I(inode);
531 	struct super_block *sb = inode->i_sb;
532 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
533 	struct nilfs_inode *raw_inode;
534 
535 	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
536 
537 	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
538 		memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
539 	set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
540 
541 	nilfs_write_inode_common(inode, raw_inode, 0);
542 		/* XXX: call with has_bmap = 0 is a workaround to avoid
543 		   deadlock of bmap. This delays update of i_bmap to just
544 		   before writing */
545 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
546 }
547 
548 #define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
549 
550 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
551 				unsigned long from)
552 {
553 	unsigned long b;
554 	int ret;
555 
556 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
557 		return;
558  repeat:
559 	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
560 	if (ret == -ENOENT)
561 		return;
562 	else if (ret < 0)
563 		goto failed;
564 
565 	if (b < from)
566 		return;
567 
568 	b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
569 	ret = nilfs_bmap_truncate(ii->i_bmap, b);
570 	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
571 	if (!ret || (ret == -ENOMEM &&
572 		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
573 		goto repeat;
574 
575  failed:
576 	if (ret == -EINVAL)
577 		nilfs_error(ii->vfs_inode.i_sb, __func__,
578 			    "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
579 	else
580 		nilfs_warning(ii->vfs_inode.i_sb, __func__,
581 			      "failed to truncate bmap (ino=%lu, err=%d)",
582 			      ii->vfs_inode.i_ino, ret);
583 }
584 
585 void nilfs_truncate(struct inode *inode)
586 {
587 	unsigned long blkoff;
588 	unsigned int blocksize;
589 	struct nilfs_transaction_info ti;
590 	struct super_block *sb = inode->i_sb;
591 	struct nilfs_inode_info *ii = NILFS_I(inode);
592 
593 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
594 		return;
595 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
596 		return;
597 
598 	blocksize = sb->s_blocksize;
599 	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
600 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
601 
602 	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
603 
604 	nilfs_truncate_bmap(ii, blkoff);
605 
606 	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
607 	if (IS_SYNC(inode))
608 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
609 
610 	nilfs_mark_inode_dirty(inode);
611 	nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
612 	nilfs_transaction_commit(sb);
613 	/* May construct a logical segment and may fail in sync mode.
614 	   But truncate has no return value. */
615 }
616 
617 void nilfs_delete_inode(struct inode *inode)
618 {
619 	struct nilfs_transaction_info ti;
620 	struct super_block *sb = inode->i_sb;
621 	struct nilfs_inode_info *ii = NILFS_I(inode);
622 
623 	if (unlikely(is_bad_inode(inode))) {
624 		if (inode->i_data.nrpages)
625 			truncate_inode_pages(&inode->i_data, 0);
626 		clear_inode(inode);
627 		return;
628 	}
629 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
630 
631 	if (inode->i_data.nrpages)
632 		truncate_inode_pages(&inode->i_data, 0);
633 
634 	nilfs_truncate_bmap(ii, 0);
635 	nilfs_mark_inode_dirty(inode);
636 	nilfs_free_inode(inode);
637 	/* nilfs_free_inode() marks inode buffer dirty */
638 	if (IS_SYNC(inode))
639 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
640 	nilfs_transaction_commit(sb);
641 	/* May construct a logical segment and may fail in sync mode.
642 	   But delete_inode has no return value. */
643 }
644 
645 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
646 {
647 	struct nilfs_transaction_info ti;
648 	struct inode *inode = dentry->d_inode;
649 	struct super_block *sb = inode->i_sb;
650 	int err;
651 
652 	err = inode_change_ok(inode, iattr);
653 	if (err)
654 		return err;
655 
656 	err = nilfs_transaction_begin(sb, &ti, 0);
657 	if (unlikely(err))
658 		return err;
659 	err = inode_setattr(inode, iattr);
660 	if (!err && (iattr->ia_valid & ATTR_MODE))
661 		err = nilfs_acl_chmod(inode);
662 	if (likely(!err))
663 		err = nilfs_transaction_commit(sb);
664 	else
665 		nilfs_transaction_abort(sb);
666 
667 	return err;
668 }
669 
670 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
671 			   struct buffer_head **pbh)
672 {
673 	struct nilfs_inode_info *ii = NILFS_I(inode);
674 	int err;
675 
676 	spin_lock(&sbi->s_inode_lock);
677 	if (ii->i_bh == NULL) {
678 		spin_unlock(&sbi->s_inode_lock);
679 		err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
680 						  pbh);
681 		if (unlikely(err))
682 			return err;
683 		spin_lock(&sbi->s_inode_lock);
684 		if (ii->i_bh == NULL)
685 			ii->i_bh = *pbh;
686 		else {
687 			brelse(*pbh);
688 			*pbh = ii->i_bh;
689 		}
690 	} else
691 		*pbh = ii->i_bh;
692 
693 	get_bh(*pbh);
694 	spin_unlock(&sbi->s_inode_lock);
695 	return 0;
696 }
697 
698 int nilfs_inode_dirty(struct inode *inode)
699 {
700 	struct nilfs_inode_info *ii = NILFS_I(inode);
701 	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
702 	int ret = 0;
703 
704 	if (!list_empty(&ii->i_dirty)) {
705 		spin_lock(&sbi->s_inode_lock);
706 		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
707 			test_bit(NILFS_I_BUSY, &ii->i_state);
708 		spin_unlock(&sbi->s_inode_lock);
709 	}
710 	return ret;
711 }
712 
713 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
714 			 unsigned nr_dirty)
715 {
716 	struct nilfs_inode_info *ii = NILFS_I(inode);
717 
718 	atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
719 
720 	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
721 		return 0;
722 
723 	spin_lock(&sbi->s_inode_lock);
724 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
725 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
726 		/* Because this routine may race with nilfs_dispose_list(),
727 		   we have to check NILFS_I_QUEUED here, too. */
728 		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
729 			/* This will happen when somebody is freeing
730 			   this inode. */
731 			nilfs_warning(sbi->s_super, __func__,
732 				      "cannot get inode (ino=%lu)\n",
733 				      inode->i_ino);
734 			spin_unlock(&sbi->s_inode_lock);
735 			return -EINVAL; /* NILFS_I_DIRTY may remain for
736 					   freeing inode */
737 		}
738 		list_del(&ii->i_dirty);
739 		list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
740 		set_bit(NILFS_I_QUEUED, &ii->i_state);
741 	}
742 	spin_unlock(&sbi->s_inode_lock);
743 	return 0;
744 }
745 
746 int nilfs_mark_inode_dirty(struct inode *inode)
747 {
748 	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
749 	struct buffer_head *ibh;
750 	int err;
751 
752 	err = nilfs_load_inode_block(sbi, inode, &ibh);
753 	if (unlikely(err)) {
754 		nilfs_warning(inode->i_sb, __func__,
755 			      "failed to reget inode block.\n");
756 		return err;
757 	}
758 	nilfs_update_inode(inode, ibh);
759 	nilfs_mdt_mark_buffer_dirty(ibh);
760 	nilfs_mdt_mark_dirty(sbi->s_ifile);
761 	brelse(ibh);
762 	return 0;
763 }
764 
765 /**
766  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
767  * @inode: inode of the file to be registered.
768  *
769  * nilfs_dirty_inode() loads a inode block containing the specified
770  * @inode and copies data from a nilfs_inode to a corresponding inode
771  * entry in the inode block. This operation is excluded from the segment
772  * construction. This function can be called both as a single operation
773  * and as a part of indivisible file operations.
774  */
775 void nilfs_dirty_inode(struct inode *inode)
776 {
777 	struct nilfs_transaction_info ti;
778 
779 	if (is_bad_inode(inode)) {
780 		nilfs_warning(inode->i_sb, __func__,
781 			      "tried to mark bad_inode dirty. ignored.\n");
782 		dump_stack();
783 		return;
784 	}
785 	nilfs_transaction_begin(inode->i_sb, &ti, 0);
786 	nilfs_mark_inode_dirty(inode);
787 	nilfs_transaction_commit(inode->i_sb); /* never fails */
788 }
789