inode.c revision abdb318b79d387a723af5db2aa79f812cefd0797
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/mpage.h>
26#include <linux/writeback.h>
27#include <linux/uio.h>
28#include "nilfs.h"
29#include "segment.h"
30#include "page.h"
31#include "mdt.h"
32#include "cpfile.h"
33#include "ifile.h"
34
35
36/**
37 * nilfs_get_block() - get a file block on the filesystem (callback function)
38 * @inode - inode struct of the target file
39 * @blkoff - file block number
40 * @bh_result - buffer head to be mapped on
41 * @create - indicate whether allocating the block or not when it has not
42 *      been allocated yet.
43 *
44 * This function does not issue actual read request of the specified data
45 * block. It is done by VFS.
46 */
47int nilfs_get_block(struct inode *inode, sector_t blkoff,
48		    struct buffer_head *bh_result, int create)
49{
50	struct nilfs_inode_info *ii = NILFS_I(inode);
51	__u64 blknum = 0;
52	int err = 0, ret;
53	struct inode *dat = nilfs_dat_inode(NILFS_I_NILFS(inode));
54	unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
55
56	down_read(&NILFS_MDT(dat)->mi_sem);
57	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
58	up_read(&NILFS_MDT(dat)->mi_sem);
59	if (ret >= 0) {	/* found */
60		map_bh(bh_result, inode->i_sb, blknum);
61		if (ret > 0)
62			bh_result->b_size = (ret << inode->i_blkbits);
63		goto out;
64	}
65	/* data block was not found */
66	if (ret == -ENOENT && create) {
67		struct nilfs_transaction_info ti;
68
69		bh_result->b_blocknr = 0;
70		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
71		if (unlikely(err))
72			goto out;
73		err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
74					(unsigned long)bh_result);
75		if (unlikely(err != 0)) {
76			if (err == -EEXIST) {
77				/*
78				 * The get_block() function could be called
79				 * from multiple callers for an inode.
80				 * However, the page having this block must
81				 * be locked in this case.
82				 */
83				printk(KERN_WARNING
84				       "nilfs_get_block: a race condition "
85				       "while inserting a data block. "
86				       "(inode number=%lu, file block "
87				       "offset=%llu)\n",
88				       inode->i_ino,
89				       (unsigned long long)blkoff);
90				err = 0;
91			} else if (err == -EINVAL) {
92				nilfs_error(inode->i_sb, __func__,
93					    "broken bmap (inode=%lu)\n",
94					    inode->i_ino);
95				err = -EIO;
96			}
97			nilfs_transaction_abort(inode->i_sb);
98			goto out;
99		}
100		nilfs_mark_inode_dirty(inode);
101		nilfs_transaction_commit(inode->i_sb); /* never fails */
102		/* Error handling should be detailed */
103		set_buffer_new(bh_result);
104		map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
105						      to proper value */
106	} else if (ret == -ENOENT) {
107		/* not found is not error (e.g. hole); must return without
108		   the mapped state flag. */
109		;
110	} else {
111		err = ret;
112	}
113
114 out:
115	return err;
116}
117
118/**
119 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
120 * address_space_operations.
121 * @file - file struct of the file to be read
122 * @page - the page to be read
123 */
124static int nilfs_readpage(struct file *file, struct page *page)
125{
126	return mpage_readpage(page, nilfs_get_block);
127}
128
129/**
130 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
131 * address_space_operations.
132 * @file - file struct of the file to be read
133 * @mapping - address_space struct used for reading multiple pages
134 * @pages - the pages to be read
135 * @nr_pages - number of pages to be read
136 */
137static int nilfs_readpages(struct file *file, struct address_space *mapping,
138			   struct list_head *pages, unsigned nr_pages)
139{
140	return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
141}
142
143static int nilfs_writepages(struct address_space *mapping,
144			    struct writeback_control *wbc)
145{
146	struct inode *inode = mapping->host;
147	int err = 0;
148
149	if (wbc->sync_mode == WB_SYNC_ALL)
150		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
151						    wbc->range_start,
152						    wbc->range_end);
153	return err;
154}
155
156static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
157{
158	struct inode *inode = page->mapping->host;
159	int err;
160
161	redirty_page_for_writepage(wbc, page);
162	unlock_page(page);
163
164	if (wbc->sync_mode == WB_SYNC_ALL) {
165		err = nilfs_construct_segment(inode->i_sb);
166		if (unlikely(err))
167			return err;
168	} else if (wbc->for_reclaim)
169		nilfs_flush_segment(inode->i_sb, inode->i_ino);
170
171	return 0;
172}
173
174static int nilfs_set_page_dirty(struct page *page)
175{
176	int ret = __set_page_dirty_buffers(page);
177
178	if (ret) {
179		struct inode *inode = page->mapping->host;
180		struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
181		unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
182
183		nilfs_set_file_dirty(sbi, inode, nr_dirty);
184	}
185	return ret;
186}
187
188static int nilfs_write_begin(struct file *file, struct address_space *mapping,
189			     loff_t pos, unsigned len, unsigned flags,
190			     struct page **pagep, void **fsdata)
191
192{
193	struct inode *inode = mapping->host;
194	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
195
196	if (unlikely(err))
197		return err;
198
199	*pagep = NULL;
200	err = block_write_begin(file, mapping, pos, len, flags, pagep,
201				fsdata, nilfs_get_block);
202	if (unlikely(err))
203		nilfs_transaction_abort(inode->i_sb);
204	return err;
205}
206
207static int nilfs_write_end(struct file *file, struct address_space *mapping,
208			   loff_t pos, unsigned len, unsigned copied,
209			   struct page *page, void *fsdata)
210{
211	struct inode *inode = mapping->host;
212	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
213	unsigned nr_dirty;
214	int err;
215
216	nr_dirty = nilfs_page_count_clean_buffers(page, start,
217						  start + copied);
218	copied = generic_write_end(file, mapping, pos, len, copied, page,
219				   fsdata);
220	nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
221	err = nilfs_transaction_commit(inode->i_sb);
222	return err ? : copied;
223}
224
225static ssize_t
226nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
227		loff_t offset, unsigned long nr_segs)
228{
229	struct file *file = iocb->ki_filp;
230	struct inode *inode = file->f_mapping->host;
231	ssize_t size;
232
233	if (rw == WRITE)
234		return 0;
235
236	/* Needs synchronization with the cleaner */
237	size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
238				  offset, nr_segs, nilfs_get_block, NULL);
239	return size;
240}
241
242const struct address_space_operations nilfs_aops = {
243	.writepage		= nilfs_writepage,
244	.readpage		= nilfs_readpage,
245	.sync_page		= block_sync_page,
246	.writepages		= nilfs_writepages,
247	.set_page_dirty		= nilfs_set_page_dirty,
248	.readpages		= nilfs_readpages,
249	.write_begin		= nilfs_write_begin,
250	.write_end		= nilfs_write_end,
251	/* .releasepage		= nilfs_releasepage, */
252	.invalidatepage		= block_invalidatepage,
253	.direct_IO		= nilfs_direct_IO,
254	.is_partially_uptodate  = block_is_partially_uptodate,
255};
256
257struct inode *nilfs_new_inode(struct inode *dir, int mode)
258{
259	struct super_block *sb = dir->i_sb;
260	struct nilfs_sb_info *sbi = NILFS_SB(sb);
261	struct inode *inode;
262	struct nilfs_inode_info *ii;
263	int err = -ENOMEM;
264	ino_t ino;
265
266	inode = new_inode(sb);
267	if (unlikely(!inode))
268		goto failed;
269
270	mapping_set_gfp_mask(inode->i_mapping,
271			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
272
273	ii = NILFS_I(inode);
274	ii->i_state = 1 << NILFS_I_NEW;
275
276	err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
277	if (unlikely(err))
278		goto failed_ifile_create_inode;
279	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
280
281	atomic_inc(&sbi->s_inodes_count);
282
283	inode->i_uid = current_fsuid();
284	if (dir->i_mode & S_ISGID) {
285		inode->i_gid = dir->i_gid;
286		if (S_ISDIR(mode))
287			mode |= S_ISGID;
288	} else
289		inode->i_gid = current_fsgid();
290
291	inode->i_mode = mode;
292	inode->i_ino = ino;
293	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
294
295	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
296		err = nilfs_bmap_read(ii->i_bmap, NULL);
297		if (err < 0)
298			goto failed_bmap;
299
300		set_bit(NILFS_I_BMAP, &ii->i_state);
301		/* No lock is needed; iget() ensures it. */
302	}
303
304	ii->i_flags = NILFS_I(dir)->i_flags;
305	if (S_ISLNK(mode))
306		ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
307	if (!S_ISDIR(mode))
308		ii->i_flags &= ~NILFS_DIRSYNC_FL;
309
310	/* ii->i_file_acl = 0; */
311	/* ii->i_dir_acl = 0; */
312	ii->i_dir_start_lookup = 0;
313	ii->i_cno = 0;
314	nilfs_set_inode_flags(inode);
315	spin_lock(&sbi->s_next_gen_lock);
316	inode->i_generation = sbi->s_next_generation++;
317	spin_unlock(&sbi->s_next_gen_lock);
318	insert_inode_hash(inode);
319
320	err = nilfs_init_acl(inode, dir);
321	if (unlikely(err))
322		goto failed_acl; /* never occur. When supporting
323				    nilfs_init_acl(), proper cancellation of
324				    above jobs should be considered */
325
326	return inode;
327
328 failed_acl:
329 failed_bmap:
330	inode->i_nlink = 0;
331	iput(inode);  /* raw_inode will be deleted through
332			 generic_delete_inode() */
333	goto failed;
334
335 failed_ifile_create_inode:
336	make_bad_inode(inode);
337	iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
338			 called */
339 failed:
340	return ERR_PTR(err);
341}
342
343void nilfs_free_inode(struct inode *inode)
344{
345	struct super_block *sb = inode->i_sb;
346	struct nilfs_sb_info *sbi = NILFS_SB(sb);
347
348	clear_inode(inode);
349	/* XXX: check error code? Is there any thing I can do? */
350	(void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
351	atomic_dec(&sbi->s_inodes_count);
352}
353
354void nilfs_set_inode_flags(struct inode *inode)
355{
356	unsigned int flags = NILFS_I(inode)->i_flags;
357
358	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
359			    S_DIRSYNC);
360	if (flags & NILFS_SYNC_FL)
361		inode->i_flags |= S_SYNC;
362	if (flags & NILFS_APPEND_FL)
363		inode->i_flags |= S_APPEND;
364	if (flags & NILFS_IMMUTABLE_FL)
365		inode->i_flags |= S_IMMUTABLE;
366#ifndef NILFS_ATIME_DISABLE
367	if (flags & NILFS_NOATIME_FL)
368#endif
369		inode->i_flags |= S_NOATIME;
370	if (flags & NILFS_DIRSYNC_FL)
371		inode->i_flags |= S_DIRSYNC;
372	mapping_set_gfp_mask(inode->i_mapping,
373			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
374}
375
376int nilfs_read_inode_common(struct inode *inode,
377			    struct nilfs_inode *raw_inode)
378{
379	struct nilfs_inode_info *ii = NILFS_I(inode);
380	int err;
381
382	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
383	inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
384	inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
385	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
386	inode->i_size = le64_to_cpu(raw_inode->i_size);
387	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
388	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
389	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
390	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
391	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
392	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
393	if (inode->i_nlink == 0 && inode->i_mode == 0)
394		return -EINVAL; /* this inode is deleted */
395
396	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
397	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
398#if 0
399	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
400	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
401		0 : le32_to_cpu(raw_inode->i_dir_acl);
402#endif
403	ii->i_dir_start_lookup = 0;
404	ii->i_cno = 0;
405	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
406
407	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
408	    S_ISLNK(inode->i_mode)) {
409		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
410		if (err < 0)
411			return err;
412		set_bit(NILFS_I_BMAP, &ii->i_state);
413		/* No lock is needed; iget() ensures it. */
414	}
415	return 0;
416}
417
418static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
419			      struct inode *inode)
420{
421	struct nilfs_sb_info *sbi = NILFS_SB(sb);
422	struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
423	struct buffer_head *bh;
424	struct nilfs_inode *raw_inode;
425	int err;
426
427	down_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
428	err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
429	if (unlikely(err))
430		goto bad_inode;
431
432	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
433
434	err = nilfs_read_inode_common(inode, raw_inode);
435	if (err)
436		goto failed_unmap;
437
438	if (S_ISREG(inode->i_mode)) {
439		inode->i_op = &nilfs_file_inode_operations;
440		inode->i_fop = &nilfs_file_operations;
441		inode->i_mapping->a_ops = &nilfs_aops;
442	} else if (S_ISDIR(inode->i_mode)) {
443		inode->i_op = &nilfs_dir_inode_operations;
444		inode->i_fop = &nilfs_dir_operations;
445		inode->i_mapping->a_ops = &nilfs_aops;
446	} else if (S_ISLNK(inode->i_mode)) {
447		inode->i_op = &nilfs_symlink_inode_operations;
448		inode->i_mapping->a_ops = &nilfs_aops;
449	} else {
450		inode->i_op = &nilfs_special_inode_operations;
451		init_special_inode(
452			inode, inode->i_mode,
453			new_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
454	}
455	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
456	brelse(bh);
457	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
458	nilfs_set_inode_flags(inode);
459	return 0;
460
461 failed_unmap:
462	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
463	brelse(bh);
464
465 bad_inode:
466	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
467	return err;
468}
469
470struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
471{
472	struct inode *inode;
473	int err;
474
475	inode = iget_locked(sb, ino);
476	if (unlikely(!inode))
477		return ERR_PTR(-ENOMEM);
478	if (!(inode->i_state & I_NEW))
479		return inode;
480
481	err = __nilfs_read_inode(sb, ino, inode);
482	if (unlikely(err)) {
483		iget_failed(inode);
484		return ERR_PTR(err);
485	}
486	unlock_new_inode(inode);
487	return inode;
488}
489
490void nilfs_write_inode_common(struct inode *inode,
491			      struct nilfs_inode *raw_inode, int has_bmap)
492{
493	struct nilfs_inode_info *ii = NILFS_I(inode);
494
495	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
496	raw_inode->i_uid = cpu_to_le32(inode->i_uid);
497	raw_inode->i_gid = cpu_to_le32(inode->i_gid);
498	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
499	raw_inode->i_size = cpu_to_le64(inode->i_size);
500	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
501	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
502	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
503	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
504	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
505
506	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
507	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
508
509	if (has_bmap)
510		nilfs_bmap_write(ii->i_bmap, raw_inode);
511	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
512		raw_inode->i_device_code =
513			cpu_to_le64(new_encode_dev(inode->i_rdev));
514	/* When extending inode, nilfs->ns_inode_size should be checked
515	   for substitutions of appended fields */
516}
517
518void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
519{
520	ino_t ino = inode->i_ino;
521	struct nilfs_inode_info *ii = NILFS_I(inode);
522	struct super_block *sb = inode->i_sb;
523	struct nilfs_sb_info *sbi = NILFS_SB(sb);
524	struct nilfs_inode *raw_inode;
525
526	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
527
528	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
529		memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
530	set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
531
532	nilfs_write_inode_common(inode, raw_inode, 0);
533		/* XXX: call with has_bmap = 0 is a workaround to avoid
534		   deadlock of bmap. This delays update of i_bmap to just
535		   before writing */
536	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
537}
538
539#define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
540
541static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
542				unsigned long from)
543{
544	unsigned long b;
545	int ret;
546
547	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
548		return;
549 repeat:
550	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
551	if (ret == -ENOENT)
552		return;
553	else if (ret < 0)
554		goto failed;
555
556	if (b < from)
557		return;
558
559	b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
560	ret = nilfs_bmap_truncate(ii->i_bmap, b);
561	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
562	if (!ret || (ret == -ENOMEM &&
563		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
564		goto repeat;
565
566 failed:
567	if (ret == -EINVAL)
568		nilfs_error(ii->vfs_inode.i_sb, __func__,
569			    "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
570	else
571		nilfs_warning(ii->vfs_inode.i_sb, __func__,
572			      "failed to truncate bmap (ino=%lu, err=%d)",
573			      ii->vfs_inode.i_ino, ret);
574}
575
576void nilfs_truncate(struct inode *inode)
577{
578	unsigned long blkoff;
579	unsigned int blocksize;
580	struct nilfs_transaction_info ti;
581	struct super_block *sb = inode->i_sb;
582	struct nilfs_inode_info *ii = NILFS_I(inode);
583
584	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
585		return;
586	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
587		return;
588
589	blocksize = sb->s_blocksize;
590	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
591	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
592
593	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
594
595	nilfs_truncate_bmap(ii, blkoff);
596
597	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
598	if (IS_SYNC(inode))
599		nilfs_set_transaction_flag(NILFS_TI_SYNC);
600
601	nilfs_mark_inode_dirty(inode);
602	nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
603	nilfs_transaction_commit(sb);
604	/* May construct a logical segment and may fail in sync mode.
605	   But truncate has no return value. */
606}
607
608void nilfs_delete_inode(struct inode *inode)
609{
610	struct nilfs_transaction_info ti;
611	struct super_block *sb = inode->i_sb;
612	struct nilfs_inode_info *ii = NILFS_I(inode);
613
614	if (unlikely(is_bad_inode(inode))) {
615		if (inode->i_data.nrpages)
616			truncate_inode_pages(&inode->i_data, 0);
617		clear_inode(inode);
618		return;
619	}
620	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
621
622	if (inode->i_data.nrpages)
623		truncate_inode_pages(&inode->i_data, 0);
624
625	nilfs_truncate_bmap(ii, 0);
626	nilfs_mark_inode_dirty(inode);
627	nilfs_free_inode(inode);
628	/* nilfs_free_inode() marks inode buffer dirty */
629	if (IS_SYNC(inode))
630		nilfs_set_transaction_flag(NILFS_TI_SYNC);
631	nilfs_transaction_commit(sb);
632	/* May construct a logical segment and may fail in sync mode.
633	   But delete_inode has no return value. */
634}
635
636int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
637{
638	struct nilfs_transaction_info ti;
639	struct inode *inode = dentry->d_inode;
640	struct super_block *sb = inode->i_sb;
641	int err;
642
643	err = inode_change_ok(inode, iattr);
644	if (err)
645		return err;
646
647	err = nilfs_transaction_begin(sb, &ti, 0);
648	if (unlikely(err))
649		return err;
650	err = inode_setattr(inode, iattr);
651	if (!err && (iattr->ia_valid & ATTR_MODE))
652		err = nilfs_acl_chmod(inode);
653	if (likely(!err))
654		err = nilfs_transaction_commit(sb);
655	else
656		nilfs_transaction_abort(sb);
657
658	return err;
659}
660
661int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
662			   struct buffer_head **pbh)
663{
664	struct nilfs_inode_info *ii = NILFS_I(inode);
665	int err;
666
667	spin_lock(&sbi->s_inode_lock);
668	if (ii->i_bh == NULL) {
669		spin_unlock(&sbi->s_inode_lock);
670		err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
671						  pbh);
672		if (unlikely(err))
673			return err;
674		spin_lock(&sbi->s_inode_lock);
675		if (ii->i_bh == NULL)
676			ii->i_bh = *pbh;
677		else {
678			brelse(*pbh);
679			*pbh = ii->i_bh;
680		}
681	} else
682		*pbh = ii->i_bh;
683
684	get_bh(*pbh);
685	spin_unlock(&sbi->s_inode_lock);
686	return 0;
687}
688
689int nilfs_inode_dirty(struct inode *inode)
690{
691	struct nilfs_inode_info *ii = NILFS_I(inode);
692	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
693	int ret = 0;
694
695	if (!list_empty(&ii->i_dirty)) {
696		spin_lock(&sbi->s_inode_lock);
697		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
698			test_bit(NILFS_I_BUSY, &ii->i_state);
699		spin_unlock(&sbi->s_inode_lock);
700	}
701	return ret;
702}
703
704int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
705			 unsigned nr_dirty)
706{
707	struct nilfs_inode_info *ii = NILFS_I(inode);
708
709	atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
710
711	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
712		return 0;
713
714	spin_lock(&sbi->s_inode_lock);
715	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
716	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
717		/* Because this routine may race with nilfs_dispose_list(),
718		   we have to check NILFS_I_QUEUED here, too. */
719		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
720			/* This will happen when somebody is freeing
721			   this inode. */
722			nilfs_warning(sbi->s_super, __func__,
723				      "cannot get inode (ino=%lu)\n",
724				      inode->i_ino);
725			spin_unlock(&sbi->s_inode_lock);
726			return -EINVAL; /* NILFS_I_DIRTY may remain for
727					   freeing inode */
728		}
729		list_del(&ii->i_dirty);
730		list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
731		set_bit(NILFS_I_QUEUED, &ii->i_state);
732	}
733	spin_unlock(&sbi->s_inode_lock);
734	return 0;
735}
736
737int nilfs_mark_inode_dirty(struct inode *inode)
738{
739	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
740	struct buffer_head *ibh;
741	int err;
742
743	err = nilfs_load_inode_block(sbi, inode, &ibh);
744	if (unlikely(err)) {
745		nilfs_warning(inode->i_sb, __func__,
746			      "failed to reget inode block.\n");
747		return err;
748	}
749	nilfs_update_inode(inode, ibh);
750	nilfs_mdt_mark_buffer_dirty(ibh);
751	nilfs_mdt_mark_dirty(sbi->s_ifile);
752	brelse(ibh);
753	return 0;
754}
755
756/**
757 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
758 * @inode: inode of the file to be registered.
759 *
760 * nilfs_dirty_inode() loads a inode block containing the specified
761 * @inode and copies data from a nilfs_inode to a corresponding inode
762 * entry in the inode block. This operation is excluded from the segment
763 * construction. This function can be called both as a single operation
764 * and as a part of indivisible file operations.
765 */
766void nilfs_dirty_inode(struct inode *inode)
767{
768	struct nilfs_transaction_info ti;
769
770	if (is_bad_inode(inode)) {
771		nilfs_warning(inode->i_sb, __func__,
772			      "tried to mark bad_inode dirty. ignored.\n");
773		dump_stack();
774		return;
775	}
776	nilfs_transaction_begin(inode->i_sb, &ti, 0);
777	nilfs_mark_inode_dirty(inode);
778	nilfs_transaction_commit(inode->i_sb); /* never fails */
779}
780