xref: /openbmc/linux/fs/nilfs2/inode.c (revision 5d4a2e29)
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 	*pagep = NULL;
201 	err = block_write_begin(file, mapping, pos, len, flags, pagep,
202 				fsdata, nilfs_get_block);
203 	if (unlikely(err))
204 		nilfs_transaction_abort(inode->i_sb);
205 	return err;
206 }
207 
208 static int nilfs_write_end(struct file *file, struct address_space *mapping,
209 			   loff_t pos, unsigned len, unsigned copied,
210 			   struct page *page, void *fsdata)
211 {
212 	struct inode *inode = mapping->host;
213 	unsigned start = pos & (PAGE_CACHE_SIZE - 1);
214 	unsigned nr_dirty;
215 	int err;
216 
217 	nr_dirty = nilfs_page_count_clean_buffers(page, start,
218 						  start + copied);
219 	copied = generic_write_end(file, mapping, pos, len, copied, page,
220 				   fsdata);
221 	nilfs_set_file_dirty(NILFS_SB(inode->i_sb), inode, nr_dirty);
222 	err = nilfs_transaction_commit(inode->i_sb);
223 	return err ? : copied;
224 }
225 
226 static ssize_t
227 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
228 		loff_t offset, unsigned long nr_segs)
229 {
230 	struct file *file = iocb->ki_filp;
231 	struct inode *inode = file->f_mapping->host;
232 	ssize_t size;
233 
234 	if (rw == WRITE)
235 		return 0;
236 
237 	/* Needs synchronization with the cleaner */
238 	size = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
239 				  offset, nr_segs, nilfs_get_block, NULL);
240 	return size;
241 }
242 
243 const struct address_space_operations nilfs_aops = {
244 	.writepage		= nilfs_writepage,
245 	.readpage		= nilfs_readpage,
246 	.sync_page		= block_sync_page,
247 	.writepages		= nilfs_writepages,
248 	.set_page_dirty		= nilfs_set_page_dirty,
249 	.readpages		= nilfs_readpages,
250 	.write_begin		= nilfs_write_begin,
251 	.write_end		= nilfs_write_end,
252 	/* .releasepage		= nilfs_releasepage, */
253 	.invalidatepage		= block_invalidatepage,
254 	.direct_IO		= nilfs_direct_IO,
255 	.is_partially_uptodate  = block_is_partially_uptodate,
256 };
257 
258 struct inode *nilfs_new_inode(struct inode *dir, int mode)
259 {
260 	struct super_block *sb = dir->i_sb;
261 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
262 	struct inode *inode;
263 	struct nilfs_inode_info *ii;
264 	int err = -ENOMEM;
265 	ino_t ino;
266 
267 	inode = new_inode(sb);
268 	if (unlikely(!inode))
269 		goto failed;
270 
271 	mapping_set_gfp_mask(inode->i_mapping,
272 			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
273 
274 	ii = NILFS_I(inode);
275 	ii->i_state = 1 << NILFS_I_NEW;
276 
277 	err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh);
278 	if (unlikely(err))
279 		goto failed_ifile_create_inode;
280 	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
281 
282 	atomic_inc(&sbi->s_inodes_count);
283 	inode_init_owner(inode, dir, mode);
284 	inode->i_ino = ino;
285 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
286 
287 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
288 		err = nilfs_bmap_read(ii->i_bmap, NULL);
289 		if (err < 0)
290 			goto failed_bmap;
291 
292 		set_bit(NILFS_I_BMAP, &ii->i_state);
293 		/* No lock is needed; iget() ensures it. */
294 	}
295 
296 	ii->i_flags = NILFS_I(dir)->i_flags;
297 	if (S_ISLNK(mode))
298 		ii->i_flags &= ~(NILFS_IMMUTABLE_FL | NILFS_APPEND_FL);
299 	if (!S_ISDIR(mode))
300 		ii->i_flags &= ~NILFS_DIRSYNC_FL;
301 
302 	/* ii->i_file_acl = 0; */
303 	/* ii->i_dir_acl = 0; */
304 	ii->i_dir_start_lookup = 0;
305 	ii->i_cno = 0;
306 	nilfs_set_inode_flags(inode);
307 	spin_lock(&sbi->s_next_gen_lock);
308 	inode->i_generation = sbi->s_next_generation++;
309 	spin_unlock(&sbi->s_next_gen_lock);
310 	insert_inode_hash(inode);
311 
312 	err = nilfs_init_acl(inode, dir);
313 	if (unlikely(err))
314 		goto failed_acl; /* never occur. When supporting
315 				    nilfs_init_acl(), proper cancellation of
316 				    above jobs should be considered */
317 
318 	return inode;
319 
320  failed_acl:
321  failed_bmap:
322 	inode->i_nlink = 0;
323 	iput(inode);  /* raw_inode will be deleted through
324 			 generic_delete_inode() */
325 	goto failed;
326 
327  failed_ifile_create_inode:
328 	make_bad_inode(inode);
329 	iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
330 			 called */
331  failed:
332 	return ERR_PTR(err);
333 }
334 
335 void nilfs_free_inode(struct inode *inode)
336 {
337 	struct super_block *sb = inode->i_sb;
338 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
339 
340 	clear_inode(inode);
341 	/* XXX: check error code? Is there any thing I can do? */
342 	(void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
343 	atomic_dec(&sbi->s_inodes_count);
344 }
345 
346 void nilfs_set_inode_flags(struct inode *inode)
347 {
348 	unsigned int flags = NILFS_I(inode)->i_flags;
349 
350 	inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
351 			    S_DIRSYNC);
352 	if (flags & NILFS_SYNC_FL)
353 		inode->i_flags |= S_SYNC;
354 	if (flags & NILFS_APPEND_FL)
355 		inode->i_flags |= S_APPEND;
356 	if (flags & NILFS_IMMUTABLE_FL)
357 		inode->i_flags |= S_IMMUTABLE;
358 #ifndef NILFS_ATIME_DISABLE
359 	if (flags & NILFS_NOATIME_FL)
360 #endif
361 		inode->i_flags |= S_NOATIME;
362 	if (flags & NILFS_DIRSYNC_FL)
363 		inode->i_flags |= S_DIRSYNC;
364 	mapping_set_gfp_mask(inode->i_mapping,
365 			     mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
366 }
367 
368 int nilfs_read_inode_common(struct inode *inode,
369 			    struct nilfs_inode *raw_inode)
370 {
371 	struct nilfs_inode_info *ii = NILFS_I(inode);
372 	int err;
373 
374 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
375 	inode->i_uid = (uid_t)le32_to_cpu(raw_inode->i_uid);
376 	inode->i_gid = (gid_t)le32_to_cpu(raw_inode->i_gid);
377 	inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
378 	inode->i_size = le64_to_cpu(raw_inode->i_size);
379 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
380 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
381 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
382 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
383 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
384 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
385 	if (inode->i_nlink == 0 && inode->i_mode == 0)
386 		return -EINVAL; /* this inode is deleted */
387 
388 	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
389 	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
390 #if 0
391 	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
392 	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
393 		0 : le32_to_cpu(raw_inode->i_dir_acl);
394 #endif
395 	ii->i_dir_start_lookup = 0;
396 	ii->i_cno = 0;
397 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
398 
399 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
400 	    S_ISLNK(inode->i_mode)) {
401 		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
402 		if (err < 0)
403 			return err;
404 		set_bit(NILFS_I_BMAP, &ii->i_state);
405 		/* No lock is needed; iget() ensures it. */
406 	}
407 	return 0;
408 }
409 
410 static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
411 			      struct inode *inode)
412 {
413 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
414 	struct inode *dat = nilfs_dat_inode(sbi->s_nilfs);
415 	struct buffer_head *bh;
416 	struct nilfs_inode *raw_inode;
417 	int err;
418 
419 	down_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
420 	err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh);
421 	if (unlikely(err))
422 		goto bad_inode;
423 
424 	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh);
425 
426 	err = nilfs_read_inode_common(inode, raw_inode);
427 	if (err)
428 		goto failed_unmap;
429 
430 	if (S_ISREG(inode->i_mode)) {
431 		inode->i_op = &nilfs_file_inode_operations;
432 		inode->i_fop = &nilfs_file_operations;
433 		inode->i_mapping->a_ops = &nilfs_aops;
434 	} else if (S_ISDIR(inode->i_mode)) {
435 		inode->i_op = &nilfs_dir_inode_operations;
436 		inode->i_fop = &nilfs_dir_operations;
437 		inode->i_mapping->a_ops = &nilfs_aops;
438 	} else if (S_ISLNK(inode->i_mode)) {
439 		inode->i_op = &nilfs_symlink_inode_operations;
440 		inode->i_mapping->a_ops = &nilfs_aops;
441 	} else {
442 		inode->i_op = &nilfs_special_inode_operations;
443 		init_special_inode(
444 			inode, inode->i_mode,
445 			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
446 	}
447 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
448 	brelse(bh);
449 	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
450 	nilfs_set_inode_flags(inode);
451 	return 0;
452 
453  failed_unmap:
454 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh);
455 	brelse(bh);
456 
457  bad_inode:
458 	up_read(&NILFS_MDT(dat)->mi_sem);	/* XXX */
459 	return err;
460 }
461 
462 struct inode *nilfs_iget(struct super_block *sb, unsigned long ino)
463 {
464 	struct inode *inode;
465 	int err;
466 
467 	inode = iget_locked(sb, ino);
468 	if (unlikely(!inode))
469 		return ERR_PTR(-ENOMEM);
470 	if (!(inode->i_state & I_NEW))
471 		return inode;
472 
473 	err = __nilfs_read_inode(sb, ino, inode);
474 	if (unlikely(err)) {
475 		iget_failed(inode);
476 		return ERR_PTR(err);
477 	}
478 	unlock_new_inode(inode);
479 	return inode;
480 }
481 
482 void nilfs_write_inode_common(struct inode *inode,
483 			      struct nilfs_inode *raw_inode, int has_bmap)
484 {
485 	struct nilfs_inode_info *ii = NILFS_I(inode);
486 
487 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
488 	raw_inode->i_uid = cpu_to_le32(inode->i_uid);
489 	raw_inode->i_gid = cpu_to_le32(inode->i_gid);
490 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
491 	raw_inode->i_size = cpu_to_le64(inode->i_size);
492 	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
493 	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
494 	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
495 	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
496 	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
497 
498 	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
499 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
500 
501 	if (has_bmap)
502 		nilfs_bmap_write(ii->i_bmap, raw_inode);
503 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
504 		raw_inode->i_device_code =
505 			cpu_to_le64(huge_encode_dev(inode->i_rdev));
506 	/* When extending inode, nilfs->ns_inode_size should be checked
507 	   for substitutions of appended fields */
508 }
509 
510 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
511 {
512 	ino_t ino = inode->i_ino;
513 	struct nilfs_inode_info *ii = NILFS_I(inode);
514 	struct super_block *sb = inode->i_sb;
515 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
516 	struct nilfs_inode *raw_inode;
517 
518 	raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh);
519 
520 	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
521 		memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size);
522 	set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
523 
524 	nilfs_write_inode_common(inode, raw_inode, 0);
525 		/* XXX: call with has_bmap = 0 is a workaround to avoid
526 		   deadlock of bmap. This delays update of i_bmap to just
527 		   before writing */
528 	nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh);
529 }
530 
531 #define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
532 
533 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
534 				unsigned long from)
535 {
536 	unsigned long b;
537 	int ret;
538 
539 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
540 		return;
541  repeat:
542 	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
543 	if (ret == -ENOENT)
544 		return;
545 	else if (ret < 0)
546 		goto failed;
547 
548 	if (b < from)
549 		return;
550 
551 	b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
552 	ret = nilfs_bmap_truncate(ii->i_bmap, b);
553 	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
554 	if (!ret || (ret == -ENOMEM &&
555 		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
556 		goto repeat;
557 
558  failed:
559 	if (ret == -EINVAL)
560 		nilfs_error(ii->vfs_inode.i_sb, __func__,
561 			    "bmap is broken (ino=%lu)", ii->vfs_inode.i_ino);
562 	else
563 		nilfs_warning(ii->vfs_inode.i_sb, __func__,
564 			      "failed to truncate bmap (ino=%lu, err=%d)",
565 			      ii->vfs_inode.i_ino, ret);
566 }
567 
568 void nilfs_truncate(struct inode *inode)
569 {
570 	unsigned long blkoff;
571 	unsigned int blocksize;
572 	struct nilfs_transaction_info ti;
573 	struct super_block *sb = inode->i_sb;
574 	struct nilfs_inode_info *ii = NILFS_I(inode);
575 
576 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
577 		return;
578 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
579 		return;
580 
581 	blocksize = sb->s_blocksize;
582 	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
583 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
584 
585 	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
586 
587 	nilfs_truncate_bmap(ii, blkoff);
588 
589 	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
590 	if (IS_SYNC(inode))
591 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
592 
593 	nilfs_mark_inode_dirty(inode);
594 	nilfs_set_file_dirty(NILFS_SB(sb), inode, 0);
595 	nilfs_transaction_commit(sb);
596 	/* May construct a logical segment and may fail in sync mode.
597 	   But truncate has no return value. */
598 }
599 
600 void nilfs_delete_inode(struct inode *inode)
601 {
602 	struct nilfs_transaction_info ti;
603 	struct super_block *sb = inode->i_sb;
604 	struct nilfs_inode_info *ii = NILFS_I(inode);
605 
606 	if (unlikely(is_bad_inode(inode))) {
607 		if (inode->i_data.nrpages)
608 			truncate_inode_pages(&inode->i_data, 0);
609 		clear_inode(inode);
610 		return;
611 	}
612 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
613 
614 	if (inode->i_data.nrpages)
615 		truncate_inode_pages(&inode->i_data, 0);
616 
617 	nilfs_truncate_bmap(ii, 0);
618 	nilfs_mark_inode_dirty(inode);
619 	nilfs_free_inode(inode);
620 	/* nilfs_free_inode() marks inode buffer dirty */
621 	if (IS_SYNC(inode))
622 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
623 	nilfs_transaction_commit(sb);
624 	/* May construct a logical segment and may fail in sync mode.
625 	   But delete_inode has no return value. */
626 }
627 
628 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
629 {
630 	struct nilfs_transaction_info ti;
631 	struct inode *inode = dentry->d_inode;
632 	struct super_block *sb = inode->i_sb;
633 	int err;
634 
635 	err = inode_change_ok(inode, iattr);
636 	if (err)
637 		return err;
638 
639 	err = nilfs_transaction_begin(sb, &ti, 0);
640 	if (unlikely(err))
641 		return err;
642 	err = inode_setattr(inode, iattr);
643 	if (!err && (iattr->ia_valid & ATTR_MODE))
644 		err = nilfs_acl_chmod(inode);
645 	if (likely(!err))
646 		err = nilfs_transaction_commit(sb);
647 	else
648 		nilfs_transaction_abort(sb);
649 
650 	return err;
651 }
652 
653 int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
654 			   struct buffer_head **pbh)
655 {
656 	struct nilfs_inode_info *ii = NILFS_I(inode);
657 	int err;
658 
659 	spin_lock(&sbi->s_inode_lock);
660 	if (ii->i_bh == NULL) {
661 		spin_unlock(&sbi->s_inode_lock);
662 		err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino,
663 						  pbh);
664 		if (unlikely(err))
665 			return err;
666 		spin_lock(&sbi->s_inode_lock);
667 		if (ii->i_bh == NULL)
668 			ii->i_bh = *pbh;
669 		else {
670 			brelse(*pbh);
671 			*pbh = ii->i_bh;
672 		}
673 	} else
674 		*pbh = ii->i_bh;
675 
676 	get_bh(*pbh);
677 	spin_unlock(&sbi->s_inode_lock);
678 	return 0;
679 }
680 
681 int nilfs_inode_dirty(struct inode *inode)
682 {
683 	struct nilfs_inode_info *ii = NILFS_I(inode);
684 	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
685 	int ret = 0;
686 
687 	if (!list_empty(&ii->i_dirty)) {
688 		spin_lock(&sbi->s_inode_lock);
689 		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
690 			test_bit(NILFS_I_BUSY, &ii->i_state);
691 		spin_unlock(&sbi->s_inode_lock);
692 	}
693 	return ret;
694 }
695 
696 int nilfs_set_file_dirty(struct nilfs_sb_info *sbi, struct inode *inode,
697 			 unsigned nr_dirty)
698 {
699 	struct nilfs_inode_info *ii = NILFS_I(inode);
700 
701 	atomic_add(nr_dirty, &sbi->s_nilfs->ns_ndirtyblks);
702 
703 	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
704 		return 0;
705 
706 	spin_lock(&sbi->s_inode_lock);
707 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
708 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
709 		/* Because this routine may race with nilfs_dispose_list(),
710 		   we have to check NILFS_I_QUEUED here, too. */
711 		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
712 			/* This will happen when somebody is freeing
713 			   this inode. */
714 			nilfs_warning(sbi->s_super, __func__,
715 				      "cannot get inode (ino=%lu)\n",
716 				      inode->i_ino);
717 			spin_unlock(&sbi->s_inode_lock);
718 			return -EINVAL; /* NILFS_I_DIRTY may remain for
719 					   freeing inode */
720 		}
721 		list_del(&ii->i_dirty);
722 		list_add_tail(&ii->i_dirty, &sbi->s_dirty_files);
723 		set_bit(NILFS_I_QUEUED, &ii->i_state);
724 	}
725 	spin_unlock(&sbi->s_inode_lock);
726 	return 0;
727 }
728 
729 int nilfs_mark_inode_dirty(struct inode *inode)
730 {
731 	struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
732 	struct buffer_head *ibh;
733 	int err;
734 
735 	err = nilfs_load_inode_block(sbi, inode, &ibh);
736 	if (unlikely(err)) {
737 		nilfs_warning(inode->i_sb, __func__,
738 			      "failed to reget inode block.\n");
739 		return err;
740 	}
741 	nilfs_update_inode(inode, ibh);
742 	nilfs_mdt_mark_buffer_dirty(ibh);
743 	nilfs_mdt_mark_dirty(sbi->s_ifile);
744 	brelse(ibh);
745 	return 0;
746 }
747 
748 /**
749  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
750  * @inode: inode of the file to be registered.
751  *
752  * nilfs_dirty_inode() loads a inode block containing the specified
753  * @inode and copies data from a nilfs_inode to a corresponding inode
754  * entry in the inode block. This operation is excluded from the segment
755  * construction. This function can be called both as a single operation
756  * and as a part of indivisible file operations.
757  */
758 void nilfs_dirty_inode(struct inode *inode)
759 {
760 	struct nilfs_transaction_info ti;
761 
762 	if (is_bad_inode(inode)) {
763 		nilfs_warning(inode->i_sb, __func__,
764 			      "tried to mark bad_inode dirty. ignored.\n");
765 		dump_stack();
766 		return;
767 	}
768 	nilfs_transaction_begin(inode->i_sb, &ti, 0);
769 	nilfs_mark_inode_dirty(inode);
770 	nilfs_transaction_commit(inode->i_sb); /* never fails */
771 }
772