xref: /openbmc/linux/fs/nilfs2/inode.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * inode.c - NILFS inode operations.
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  *
9  */
10 
11 #include <linux/buffer_head.h>
12 #include <linux/gfp.h>
13 #include <linux/mpage.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/uio.h>
17 #include "nilfs.h"
18 #include "btnode.h"
19 #include "segment.h"
20 #include "page.h"
21 #include "mdt.h"
22 #include "cpfile.h"
23 #include "ifile.h"
24 
25 /**
26  * struct nilfs_iget_args - arguments used during comparison between inodes
27  * @ino: inode number
28  * @cno: checkpoint number
29  * @root: pointer on NILFS root object (mounted checkpoint)
30  * @for_gc: inode for GC flag
31  */
32 struct nilfs_iget_args {
33 	u64 ino;
34 	__u64 cno;
35 	struct nilfs_root *root;
36 	int for_gc;
37 };
38 
39 static int nilfs_iget_test(struct inode *inode, void *opaque);
40 
41 void nilfs_inode_add_blocks(struct inode *inode, int n)
42 {
43 	struct nilfs_root *root = NILFS_I(inode)->i_root;
44 
45 	inode_add_bytes(inode, i_blocksize(inode) * n);
46 	if (root)
47 		atomic64_add(n, &root->blocks_count);
48 }
49 
50 void nilfs_inode_sub_blocks(struct inode *inode, int n)
51 {
52 	struct nilfs_root *root = NILFS_I(inode)->i_root;
53 
54 	inode_sub_bytes(inode, i_blocksize(inode) * n);
55 	if (root)
56 		atomic64_sub(n, &root->blocks_count);
57 }
58 
59 /**
60  * nilfs_get_block() - get a file block on the filesystem (callback function)
61  * @inode - inode struct of the target file
62  * @blkoff - file block number
63  * @bh_result - buffer head to be mapped on
64  * @create - indicate whether allocating the block or not when it has not
65  *      been allocated yet.
66  *
67  * This function does not issue actual read request of the specified data
68  * block. It is done by VFS.
69  */
70 int nilfs_get_block(struct inode *inode, sector_t blkoff,
71 		    struct buffer_head *bh_result, int create)
72 {
73 	struct nilfs_inode_info *ii = NILFS_I(inode);
74 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
75 	__u64 blknum = 0;
76 	int err = 0, ret;
77 	unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
78 
79 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
80 	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
81 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
82 	if (ret >= 0) {	/* found */
83 		map_bh(bh_result, inode->i_sb, blknum);
84 		if (ret > 0)
85 			bh_result->b_size = (ret << inode->i_blkbits);
86 		goto out;
87 	}
88 	/* data block was not found */
89 	if (ret == -ENOENT && create) {
90 		struct nilfs_transaction_info ti;
91 
92 		bh_result->b_blocknr = 0;
93 		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
94 		if (unlikely(err))
95 			goto out;
96 		err = nilfs_bmap_insert(ii->i_bmap, blkoff,
97 					(unsigned long)bh_result);
98 		if (unlikely(err != 0)) {
99 			if (err == -EEXIST) {
100 				/*
101 				 * The get_block() function could be called
102 				 * from multiple callers for an inode.
103 				 * However, the page having this block must
104 				 * be locked in this case.
105 				 */
106 				nilfs_msg(inode->i_sb, KERN_WARNING,
107 					  "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
108 					  __func__, inode->i_ino,
109 					  (unsigned long long)blkoff);
110 				err = 0;
111 			}
112 			nilfs_transaction_abort(inode->i_sb);
113 			goto out;
114 		}
115 		nilfs_mark_inode_dirty_sync(inode);
116 		nilfs_transaction_commit(inode->i_sb); /* never fails */
117 		/* Error handling should be detailed */
118 		set_buffer_new(bh_result);
119 		set_buffer_delay(bh_result);
120 		map_bh(bh_result, inode->i_sb, 0);
121 		/* Disk block number must be changed to proper value */
122 
123 	} else if (ret == -ENOENT) {
124 		/*
125 		 * not found is not error (e.g. hole); must return without
126 		 * the mapped state flag.
127 		 */
128 		;
129 	} else {
130 		err = ret;
131 	}
132 
133  out:
134 	return err;
135 }
136 
137 /**
138  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
139  * address_space_operations.
140  * @file - file struct of the file to be read
141  * @page - the page to be read
142  */
143 static int nilfs_readpage(struct file *file, struct page *page)
144 {
145 	return mpage_readpage(page, nilfs_get_block);
146 }
147 
148 /**
149  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
150  * address_space_operations.
151  * @file - file struct of the file to be read
152  * @mapping - address_space struct used for reading multiple pages
153  * @pages - the pages to be read
154  * @nr_pages - number of pages to be read
155  */
156 static int nilfs_readpages(struct file *file, struct address_space *mapping,
157 			   struct list_head *pages, unsigned int nr_pages)
158 {
159 	return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
160 }
161 
162 static int nilfs_writepages(struct address_space *mapping,
163 			    struct writeback_control *wbc)
164 {
165 	struct inode *inode = mapping->host;
166 	int err = 0;
167 
168 	if (sb_rdonly(inode->i_sb)) {
169 		nilfs_clear_dirty_pages(mapping, false);
170 		return -EROFS;
171 	}
172 
173 	if (wbc->sync_mode == WB_SYNC_ALL)
174 		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
175 						    wbc->range_start,
176 						    wbc->range_end);
177 	return err;
178 }
179 
180 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
181 {
182 	struct inode *inode = page->mapping->host;
183 	int err;
184 
185 	if (sb_rdonly(inode->i_sb)) {
186 		/*
187 		 * It means that filesystem was remounted in read-only
188 		 * mode because of error or metadata corruption. But we
189 		 * have dirty pages that try to be flushed in background.
190 		 * So, here we simply discard this dirty page.
191 		 */
192 		nilfs_clear_dirty_page(page, false);
193 		unlock_page(page);
194 		return -EROFS;
195 	}
196 
197 	redirty_page_for_writepage(wbc, page);
198 	unlock_page(page);
199 
200 	if (wbc->sync_mode == WB_SYNC_ALL) {
201 		err = nilfs_construct_segment(inode->i_sb);
202 		if (unlikely(err))
203 			return err;
204 	} else if (wbc->for_reclaim)
205 		nilfs_flush_segment(inode->i_sb, inode->i_ino);
206 
207 	return 0;
208 }
209 
210 static int nilfs_set_page_dirty(struct page *page)
211 {
212 	struct inode *inode = page->mapping->host;
213 	int ret = __set_page_dirty_nobuffers(page);
214 
215 	if (page_has_buffers(page)) {
216 		unsigned int nr_dirty = 0;
217 		struct buffer_head *bh, *head;
218 
219 		/*
220 		 * This page is locked by callers, and no other thread
221 		 * concurrently marks its buffers dirty since they are
222 		 * only dirtied through routines in fs/buffer.c in
223 		 * which call sites of mark_buffer_dirty are protected
224 		 * by page lock.
225 		 */
226 		bh = head = page_buffers(page);
227 		do {
228 			/* Do not mark hole blocks dirty */
229 			if (buffer_dirty(bh) || !buffer_mapped(bh))
230 				continue;
231 
232 			set_buffer_dirty(bh);
233 			nr_dirty++;
234 		} while (bh = bh->b_this_page, bh != head);
235 
236 		if (nr_dirty)
237 			nilfs_set_file_dirty(inode, nr_dirty);
238 	} else if (ret) {
239 		unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
240 
241 		nilfs_set_file_dirty(inode, nr_dirty);
242 	}
243 	return ret;
244 }
245 
246 void nilfs_write_failed(struct address_space *mapping, loff_t to)
247 {
248 	struct inode *inode = mapping->host;
249 
250 	if (to > inode->i_size) {
251 		truncate_pagecache(inode, inode->i_size);
252 		nilfs_truncate(inode);
253 	}
254 }
255 
256 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
257 			     loff_t pos, unsigned len, unsigned flags,
258 			     struct page **pagep, void **fsdata)
259 
260 {
261 	struct inode *inode = mapping->host;
262 	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
263 
264 	if (unlikely(err))
265 		return err;
266 
267 	err = block_write_begin(mapping, pos, len, flags, pagep,
268 				nilfs_get_block);
269 	if (unlikely(err)) {
270 		nilfs_write_failed(mapping, pos + len);
271 		nilfs_transaction_abort(inode->i_sb);
272 	}
273 	return err;
274 }
275 
276 static int nilfs_write_end(struct file *file, struct address_space *mapping,
277 			   loff_t pos, unsigned len, unsigned copied,
278 			   struct page *page, void *fsdata)
279 {
280 	struct inode *inode = mapping->host;
281 	unsigned int start = pos & (PAGE_SIZE - 1);
282 	unsigned int nr_dirty;
283 	int err;
284 
285 	nr_dirty = nilfs_page_count_clean_buffers(page, start,
286 						  start + copied);
287 	copied = generic_write_end(file, mapping, pos, len, copied, page,
288 				   fsdata);
289 	nilfs_set_file_dirty(inode, nr_dirty);
290 	err = nilfs_transaction_commit(inode->i_sb);
291 	return err ? : copied;
292 }
293 
294 static ssize_t
295 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
296 {
297 	struct inode *inode = file_inode(iocb->ki_filp);
298 
299 	if (iov_iter_rw(iter) == WRITE)
300 		return 0;
301 
302 	/* Needs synchronization with the cleaner */
303 	return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
304 }
305 
306 const struct address_space_operations nilfs_aops = {
307 	.writepage		= nilfs_writepage,
308 	.readpage		= nilfs_readpage,
309 	.writepages		= nilfs_writepages,
310 	.set_page_dirty		= nilfs_set_page_dirty,
311 	.readpages		= nilfs_readpages,
312 	.write_begin		= nilfs_write_begin,
313 	.write_end		= nilfs_write_end,
314 	/* .releasepage		= nilfs_releasepage, */
315 	.invalidatepage		= block_invalidatepage,
316 	.direct_IO		= nilfs_direct_IO,
317 	.is_partially_uptodate  = block_is_partially_uptodate,
318 };
319 
320 static int nilfs_insert_inode_locked(struct inode *inode,
321 				     struct nilfs_root *root,
322 				     unsigned long ino)
323 {
324 	struct nilfs_iget_args args = {
325 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
326 	};
327 
328 	return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
329 }
330 
331 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
332 {
333 	struct super_block *sb = dir->i_sb;
334 	struct the_nilfs *nilfs = sb->s_fs_info;
335 	struct inode *inode;
336 	struct nilfs_inode_info *ii;
337 	struct nilfs_root *root;
338 	int err = -ENOMEM;
339 	ino_t ino;
340 
341 	inode = new_inode(sb);
342 	if (unlikely(!inode))
343 		goto failed;
344 
345 	mapping_set_gfp_mask(inode->i_mapping,
346 			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
347 
348 	root = NILFS_I(dir)->i_root;
349 	ii = NILFS_I(inode);
350 	ii->i_state = BIT(NILFS_I_NEW);
351 	ii->i_root = root;
352 
353 	err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
354 	if (unlikely(err))
355 		goto failed_ifile_create_inode;
356 	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
357 
358 	atomic64_inc(&root->inodes_count);
359 	inode_init_owner(inode, dir, mode);
360 	inode->i_ino = ino;
361 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
362 
363 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
364 		err = nilfs_bmap_read(ii->i_bmap, NULL);
365 		if (err < 0)
366 			goto failed_after_creation;
367 
368 		set_bit(NILFS_I_BMAP, &ii->i_state);
369 		/* No lock is needed; iget() ensures it. */
370 	}
371 
372 	ii->i_flags = nilfs_mask_flags(
373 		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
374 
375 	/* ii->i_file_acl = 0; */
376 	/* ii->i_dir_acl = 0; */
377 	ii->i_dir_start_lookup = 0;
378 	nilfs_set_inode_flags(inode);
379 	spin_lock(&nilfs->ns_next_gen_lock);
380 	inode->i_generation = nilfs->ns_next_generation++;
381 	spin_unlock(&nilfs->ns_next_gen_lock);
382 	if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
383 		err = -EIO;
384 		goto failed_after_creation;
385 	}
386 
387 	err = nilfs_init_acl(inode, dir);
388 	if (unlikely(err))
389 		/*
390 		 * Never occur.  When supporting nilfs_init_acl(),
391 		 * proper cancellation of above jobs should be considered.
392 		 */
393 		goto failed_after_creation;
394 
395 	return inode;
396 
397  failed_after_creation:
398 	clear_nlink(inode);
399 	unlock_new_inode(inode);
400 	iput(inode);  /*
401 		       * raw_inode will be deleted through
402 		       * nilfs_evict_inode().
403 		       */
404 	goto failed;
405 
406  failed_ifile_create_inode:
407 	make_bad_inode(inode);
408 	iput(inode);
409  failed:
410 	return ERR_PTR(err);
411 }
412 
413 void nilfs_set_inode_flags(struct inode *inode)
414 {
415 	unsigned int flags = NILFS_I(inode)->i_flags;
416 	unsigned int new_fl = 0;
417 
418 	if (flags & FS_SYNC_FL)
419 		new_fl |= S_SYNC;
420 	if (flags & FS_APPEND_FL)
421 		new_fl |= S_APPEND;
422 	if (flags & FS_IMMUTABLE_FL)
423 		new_fl |= S_IMMUTABLE;
424 	if (flags & FS_NOATIME_FL)
425 		new_fl |= S_NOATIME;
426 	if (flags & FS_DIRSYNC_FL)
427 		new_fl |= S_DIRSYNC;
428 	inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
429 			S_NOATIME | S_DIRSYNC);
430 }
431 
432 int nilfs_read_inode_common(struct inode *inode,
433 			    struct nilfs_inode *raw_inode)
434 {
435 	struct nilfs_inode_info *ii = NILFS_I(inode);
436 	int err;
437 
438 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
439 	i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
440 	i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
441 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
442 	inode->i_size = le64_to_cpu(raw_inode->i_size);
443 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
444 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
445 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
446 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
447 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
448 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
449 	if (inode->i_nlink == 0)
450 		return -ESTALE; /* this inode is deleted */
451 
452 	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
453 	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
454 #if 0
455 	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
456 	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
457 		0 : le32_to_cpu(raw_inode->i_dir_acl);
458 #endif
459 	ii->i_dir_start_lookup = 0;
460 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
461 
462 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
463 	    S_ISLNK(inode->i_mode)) {
464 		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
465 		if (err < 0)
466 			return err;
467 		set_bit(NILFS_I_BMAP, &ii->i_state);
468 		/* No lock is needed; iget() ensures it. */
469 	}
470 	return 0;
471 }
472 
473 static int __nilfs_read_inode(struct super_block *sb,
474 			      struct nilfs_root *root, unsigned long ino,
475 			      struct inode *inode)
476 {
477 	struct the_nilfs *nilfs = sb->s_fs_info;
478 	struct buffer_head *bh;
479 	struct nilfs_inode *raw_inode;
480 	int err;
481 
482 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
483 	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
484 	if (unlikely(err))
485 		goto bad_inode;
486 
487 	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
488 
489 	err = nilfs_read_inode_common(inode, raw_inode);
490 	if (err)
491 		goto failed_unmap;
492 
493 	if (S_ISREG(inode->i_mode)) {
494 		inode->i_op = &nilfs_file_inode_operations;
495 		inode->i_fop = &nilfs_file_operations;
496 		inode->i_mapping->a_ops = &nilfs_aops;
497 	} else if (S_ISDIR(inode->i_mode)) {
498 		inode->i_op = &nilfs_dir_inode_operations;
499 		inode->i_fop = &nilfs_dir_operations;
500 		inode->i_mapping->a_ops = &nilfs_aops;
501 	} else if (S_ISLNK(inode->i_mode)) {
502 		inode->i_op = &nilfs_symlink_inode_operations;
503 		inode_nohighmem(inode);
504 		inode->i_mapping->a_ops = &nilfs_aops;
505 	} else {
506 		inode->i_op = &nilfs_special_inode_operations;
507 		init_special_inode(
508 			inode, inode->i_mode,
509 			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
510 	}
511 	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
512 	brelse(bh);
513 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
514 	nilfs_set_inode_flags(inode);
515 	mapping_set_gfp_mask(inode->i_mapping,
516 			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
517 	return 0;
518 
519  failed_unmap:
520 	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
521 	brelse(bh);
522 
523  bad_inode:
524 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
525 	return err;
526 }
527 
528 static int nilfs_iget_test(struct inode *inode, void *opaque)
529 {
530 	struct nilfs_iget_args *args = opaque;
531 	struct nilfs_inode_info *ii;
532 
533 	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
534 		return 0;
535 
536 	ii = NILFS_I(inode);
537 	if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
538 		return !args->for_gc;
539 
540 	return args->for_gc && args->cno == ii->i_cno;
541 }
542 
543 static int nilfs_iget_set(struct inode *inode, void *opaque)
544 {
545 	struct nilfs_iget_args *args = opaque;
546 
547 	inode->i_ino = args->ino;
548 	if (args->for_gc) {
549 		NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
550 		NILFS_I(inode)->i_cno = args->cno;
551 		NILFS_I(inode)->i_root = NULL;
552 	} else {
553 		if (args->root && args->ino == NILFS_ROOT_INO)
554 			nilfs_get_root(args->root);
555 		NILFS_I(inode)->i_root = args->root;
556 	}
557 	return 0;
558 }
559 
560 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
561 			    unsigned long ino)
562 {
563 	struct nilfs_iget_args args = {
564 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
565 	};
566 
567 	return ilookup5(sb, ino, nilfs_iget_test, &args);
568 }
569 
570 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
571 				unsigned long ino)
572 {
573 	struct nilfs_iget_args args = {
574 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
575 	};
576 
577 	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
578 }
579 
580 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
581 			 unsigned long ino)
582 {
583 	struct inode *inode;
584 	int err;
585 
586 	inode = nilfs_iget_locked(sb, root, ino);
587 	if (unlikely(!inode))
588 		return ERR_PTR(-ENOMEM);
589 	if (!(inode->i_state & I_NEW))
590 		return inode;
591 
592 	err = __nilfs_read_inode(sb, root, ino, inode);
593 	if (unlikely(err)) {
594 		iget_failed(inode);
595 		return ERR_PTR(err);
596 	}
597 	unlock_new_inode(inode);
598 	return inode;
599 }
600 
601 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
602 				__u64 cno)
603 {
604 	struct nilfs_iget_args args = {
605 		.ino = ino, .root = NULL, .cno = cno, .for_gc = 1
606 	};
607 	struct inode *inode;
608 	int err;
609 
610 	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
611 	if (unlikely(!inode))
612 		return ERR_PTR(-ENOMEM);
613 	if (!(inode->i_state & I_NEW))
614 		return inode;
615 
616 	err = nilfs_init_gcinode(inode);
617 	if (unlikely(err)) {
618 		iget_failed(inode);
619 		return ERR_PTR(err);
620 	}
621 	unlock_new_inode(inode);
622 	return inode;
623 }
624 
625 void nilfs_write_inode_common(struct inode *inode,
626 			      struct nilfs_inode *raw_inode, int has_bmap)
627 {
628 	struct nilfs_inode_info *ii = NILFS_I(inode);
629 
630 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
631 	raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
632 	raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
633 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
634 	raw_inode->i_size = cpu_to_le64(inode->i_size);
635 	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
636 	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
637 	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
638 	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
639 	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
640 
641 	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
642 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
643 
644 	if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
645 		struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
646 
647 		/* zero-fill unused portion in the case of super root block */
648 		raw_inode->i_xattr = 0;
649 		raw_inode->i_pad = 0;
650 		memset((void *)raw_inode + sizeof(*raw_inode), 0,
651 		       nilfs->ns_inode_size - sizeof(*raw_inode));
652 	}
653 
654 	if (has_bmap)
655 		nilfs_bmap_write(ii->i_bmap, raw_inode);
656 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
657 		raw_inode->i_device_code =
658 			cpu_to_le64(huge_encode_dev(inode->i_rdev));
659 	/*
660 	 * When extending inode, nilfs->ns_inode_size should be checked
661 	 * for substitutions of appended fields.
662 	 */
663 }
664 
665 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
666 {
667 	ino_t ino = inode->i_ino;
668 	struct nilfs_inode_info *ii = NILFS_I(inode);
669 	struct inode *ifile = ii->i_root->ifile;
670 	struct nilfs_inode *raw_inode;
671 
672 	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
673 
674 	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
675 		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
676 	if (flags & I_DIRTY_DATASYNC)
677 		set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
678 
679 	nilfs_write_inode_common(inode, raw_inode, 0);
680 		/*
681 		 * XXX: call with has_bmap = 0 is a workaround to avoid
682 		 * deadlock of bmap.  This delays update of i_bmap to just
683 		 * before writing.
684 		 */
685 
686 	nilfs_ifile_unmap_inode(ifile, ino, ibh);
687 }
688 
689 #define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
690 
691 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
692 				unsigned long from)
693 {
694 	__u64 b;
695 	int ret;
696 
697 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
698 		return;
699 repeat:
700 	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
701 	if (ret == -ENOENT)
702 		return;
703 	else if (ret < 0)
704 		goto failed;
705 
706 	if (b < from)
707 		return;
708 
709 	b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
710 	ret = nilfs_bmap_truncate(ii->i_bmap, b);
711 	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
712 	if (!ret || (ret == -ENOMEM &&
713 		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
714 		goto repeat;
715 
716 failed:
717 	nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
718 		  "error %d truncating bmap (ino=%lu)", ret,
719 		  ii->vfs_inode.i_ino);
720 }
721 
722 void nilfs_truncate(struct inode *inode)
723 {
724 	unsigned long blkoff;
725 	unsigned int blocksize;
726 	struct nilfs_transaction_info ti;
727 	struct super_block *sb = inode->i_sb;
728 	struct nilfs_inode_info *ii = NILFS_I(inode);
729 
730 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
731 		return;
732 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
733 		return;
734 
735 	blocksize = sb->s_blocksize;
736 	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
737 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
738 
739 	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
740 
741 	nilfs_truncate_bmap(ii, blkoff);
742 
743 	inode->i_mtime = inode->i_ctime = current_time(inode);
744 	if (IS_SYNC(inode))
745 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
746 
747 	nilfs_mark_inode_dirty(inode);
748 	nilfs_set_file_dirty(inode, 0);
749 	nilfs_transaction_commit(sb);
750 	/*
751 	 * May construct a logical segment and may fail in sync mode.
752 	 * But truncate has no return value.
753 	 */
754 }
755 
756 static void nilfs_clear_inode(struct inode *inode)
757 {
758 	struct nilfs_inode_info *ii = NILFS_I(inode);
759 
760 	/*
761 	 * Free resources allocated in nilfs_read_inode(), here.
762 	 */
763 	BUG_ON(!list_empty(&ii->i_dirty));
764 	brelse(ii->i_bh);
765 	ii->i_bh = NULL;
766 
767 	if (nilfs_is_metadata_file_inode(inode))
768 		nilfs_mdt_clear(inode);
769 
770 	if (test_bit(NILFS_I_BMAP, &ii->i_state))
771 		nilfs_bmap_clear(ii->i_bmap);
772 
773 	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
774 
775 	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
776 		nilfs_put_root(ii->i_root);
777 }
778 
779 void nilfs_evict_inode(struct inode *inode)
780 {
781 	struct nilfs_transaction_info ti;
782 	struct super_block *sb = inode->i_sb;
783 	struct nilfs_inode_info *ii = NILFS_I(inode);
784 	int ret;
785 
786 	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
787 		truncate_inode_pages_final(&inode->i_data);
788 		clear_inode(inode);
789 		nilfs_clear_inode(inode);
790 		return;
791 	}
792 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
793 
794 	truncate_inode_pages_final(&inode->i_data);
795 
796 	/* TODO: some of the following operations may fail.  */
797 	nilfs_truncate_bmap(ii, 0);
798 	nilfs_mark_inode_dirty(inode);
799 	clear_inode(inode);
800 
801 	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
802 	if (!ret)
803 		atomic64_dec(&ii->i_root->inodes_count);
804 
805 	nilfs_clear_inode(inode);
806 
807 	if (IS_SYNC(inode))
808 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
809 	nilfs_transaction_commit(sb);
810 	/*
811 	 * May construct a logical segment and may fail in sync mode.
812 	 * But delete_inode has no return value.
813 	 */
814 }
815 
816 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
817 {
818 	struct nilfs_transaction_info ti;
819 	struct inode *inode = d_inode(dentry);
820 	struct super_block *sb = inode->i_sb;
821 	int err;
822 
823 	err = setattr_prepare(dentry, iattr);
824 	if (err)
825 		return err;
826 
827 	err = nilfs_transaction_begin(sb, &ti, 0);
828 	if (unlikely(err))
829 		return err;
830 
831 	if ((iattr->ia_valid & ATTR_SIZE) &&
832 	    iattr->ia_size != i_size_read(inode)) {
833 		inode_dio_wait(inode);
834 		truncate_setsize(inode, iattr->ia_size);
835 		nilfs_truncate(inode);
836 	}
837 
838 	setattr_copy(inode, iattr);
839 	mark_inode_dirty(inode);
840 
841 	if (iattr->ia_valid & ATTR_MODE) {
842 		err = nilfs_acl_chmod(inode);
843 		if (unlikely(err))
844 			goto out_err;
845 	}
846 
847 	return nilfs_transaction_commit(sb);
848 
849 out_err:
850 	nilfs_transaction_abort(sb);
851 	return err;
852 }
853 
854 int nilfs_permission(struct inode *inode, int mask)
855 {
856 	struct nilfs_root *root = NILFS_I(inode)->i_root;
857 
858 	if ((mask & MAY_WRITE) && root &&
859 	    root->cno != NILFS_CPTREE_CURRENT_CNO)
860 		return -EROFS; /* snapshot is not writable */
861 
862 	return generic_permission(inode, mask);
863 }
864 
865 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
866 {
867 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
868 	struct nilfs_inode_info *ii = NILFS_I(inode);
869 	int err;
870 
871 	spin_lock(&nilfs->ns_inode_lock);
872 	if (ii->i_bh == NULL) {
873 		spin_unlock(&nilfs->ns_inode_lock);
874 		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
875 						  inode->i_ino, pbh);
876 		if (unlikely(err))
877 			return err;
878 		spin_lock(&nilfs->ns_inode_lock);
879 		if (ii->i_bh == NULL)
880 			ii->i_bh = *pbh;
881 		else {
882 			brelse(*pbh);
883 			*pbh = ii->i_bh;
884 		}
885 	} else
886 		*pbh = ii->i_bh;
887 
888 	get_bh(*pbh);
889 	spin_unlock(&nilfs->ns_inode_lock);
890 	return 0;
891 }
892 
893 int nilfs_inode_dirty(struct inode *inode)
894 {
895 	struct nilfs_inode_info *ii = NILFS_I(inode);
896 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
897 	int ret = 0;
898 
899 	if (!list_empty(&ii->i_dirty)) {
900 		spin_lock(&nilfs->ns_inode_lock);
901 		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
902 			test_bit(NILFS_I_BUSY, &ii->i_state);
903 		spin_unlock(&nilfs->ns_inode_lock);
904 	}
905 	return ret;
906 }
907 
908 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
909 {
910 	struct nilfs_inode_info *ii = NILFS_I(inode);
911 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
912 
913 	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
914 
915 	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
916 		return 0;
917 
918 	spin_lock(&nilfs->ns_inode_lock);
919 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
920 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
921 		/*
922 		 * Because this routine may race with nilfs_dispose_list(),
923 		 * we have to check NILFS_I_QUEUED here, too.
924 		 */
925 		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
926 			/*
927 			 * This will happen when somebody is freeing
928 			 * this inode.
929 			 */
930 			nilfs_msg(inode->i_sb, KERN_WARNING,
931 				  "cannot set file dirty (ino=%lu): the file is being freed",
932 				  inode->i_ino);
933 			spin_unlock(&nilfs->ns_inode_lock);
934 			return -EINVAL; /*
935 					 * NILFS_I_DIRTY may remain for
936 					 * freeing inode.
937 					 */
938 		}
939 		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
940 		set_bit(NILFS_I_QUEUED, &ii->i_state);
941 	}
942 	spin_unlock(&nilfs->ns_inode_lock);
943 	return 0;
944 }
945 
946 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
947 {
948 	struct buffer_head *ibh;
949 	int err;
950 
951 	err = nilfs_load_inode_block(inode, &ibh);
952 	if (unlikely(err)) {
953 		nilfs_msg(inode->i_sb, KERN_WARNING,
954 			  "cannot mark inode dirty (ino=%lu): error %d loading inode block",
955 			  inode->i_ino, err);
956 		return err;
957 	}
958 	nilfs_update_inode(inode, ibh, flags);
959 	mark_buffer_dirty(ibh);
960 	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
961 	brelse(ibh);
962 	return 0;
963 }
964 
965 /**
966  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
967  * @inode: inode of the file to be registered.
968  *
969  * nilfs_dirty_inode() loads a inode block containing the specified
970  * @inode and copies data from a nilfs_inode to a corresponding inode
971  * entry in the inode block. This operation is excluded from the segment
972  * construction. This function can be called both as a single operation
973  * and as a part of indivisible file operations.
974  */
975 void nilfs_dirty_inode(struct inode *inode, int flags)
976 {
977 	struct nilfs_transaction_info ti;
978 	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
979 
980 	if (is_bad_inode(inode)) {
981 		nilfs_msg(inode->i_sb, KERN_WARNING,
982 			  "tried to mark bad_inode dirty. ignored.");
983 		dump_stack();
984 		return;
985 	}
986 	if (mdi) {
987 		nilfs_mdt_mark_dirty(inode);
988 		return;
989 	}
990 	nilfs_transaction_begin(inode->i_sb, &ti, 0);
991 	__nilfs_mark_inode_dirty(inode, flags);
992 	nilfs_transaction_commit(inode->i_sb); /* never fails */
993 }
994 
995 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
996 		 __u64 start, __u64 len)
997 {
998 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
999 	__u64 logical = 0, phys = 0, size = 0;
1000 	__u32 flags = 0;
1001 	loff_t isize;
1002 	sector_t blkoff, end_blkoff;
1003 	sector_t delalloc_blkoff;
1004 	unsigned long delalloc_blklen;
1005 	unsigned int blkbits = inode->i_blkbits;
1006 	int ret, n;
1007 
1008 	ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1009 	if (ret)
1010 		return ret;
1011 
1012 	inode_lock(inode);
1013 
1014 	isize = i_size_read(inode);
1015 
1016 	blkoff = start >> blkbits;
1017 	end_blkoff = (start + len - 1) >> blkbits;
1018 
1019 	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1020 							&delalloc_blkoff);
1021 
1022 	do {
1023 		__u64 blkphy;
1024 		unsigned int maxblocks;
1025 
1026 		if (delalloc_blklen && blkoff == delalloc_blkoff) {
1027 			if (size) {
1028 				/* End of the current extent */
1029 				ret = fiemap_fill_next_extent(
1030 					fieinfo, logical, phys, size, flags);
1031 				if (ret)
1032 					break;
1033 			}
1034 			if (blkoff > end_blkoff)
1035 				break;
1036 
1037 			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1038 			logical = blkoff << blkbits;
1039 			phys = 0;
1040 			size = delalloc_blklen << blkbits;
1041 
1042 			blkoff = delalloc_blkoff + delalloc_blklen;
1043 			delalloc_blklen = nilfs_find_uncommitted_extent(
1044 				inode, blkoff, &delalloc_blkoff);
1045 			continue;
1046 		}
1047 
1048 		/*
1049 		 * Limit the number of blocks that we look up so as
1050 		 * not to get into the next delayed allocation extent.
1051 		 */
1052 		maxblocks = INT_MAX;
1053 		if (delalloc_blklen)
1054 			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1055 					  maxblocks);
1056 		blkphy = 0;
1057 
1058 		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1059 		n = nilfs_bmap_lookup_contig(
1060 			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1061 		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1062 
1063 		if (n < 0) {
1064 			int past_eof;
1065 
1066 			if (unlikely(n != -ENOENT))
1067 				break; /* error */
1068 
1069 			/* HOLE */
1070 			blkoff++;
1071 			past_eof = ((blkoff << blkbits) >= isize);
1072 
1073 			if (size) {
1074 				/* End of the current extent */
1075 
1076 				if (past_eof)
1077 					flags |= FIEMAP_EXTENT_LAST;
1078 
1079 				ret = fiemap_fill_next_extent(
1080 					fieinfo, logical, phys, size, flags);
1081 				if (ret)
1082 					break;
1083 				size = 0;
1084 			}
1085 			if (blkoff > end_blkoff || past_eof)
1086 				break;
1087 		} else {
1088 			if (size) {
1089 				if (phys && blkphy << blkbits == phys + size) {
1090 					/* The current extent goes on */
1091 					size += n << blkbits;
1092 				} else {
1093 					/* Terminate the current extent */
1094 					ret = fiemap_fill_next_extent(
1095 						fieinfo, logical, phys, size,
1096 						flags);
1097 					if (ret || blkoff > end_blkoff)
1098 						break;
1099 
1100 					/* Start another extent */
1101 					flags = FIEMAP_EXTENT_MERGED;
1102 					logical = blkoff << blkbits;
1103 					phys = blkphy << blkbits;
1104 					size = n << blkbits;
1105 				}
1106 			} else {
1107 				/* Start a new extent */
1108 				flags = FIEMAP_EXTENT_MERGED;
1109 				logical = blkoff << blkbits;
1110 				phys = blkphy << blkbits;
1111 				size = n << blkbits;
1112 			}
1113 			blkoff += n;
1114 		}
1115 		cond_resched();
1116 	} while (true);
1117 
1118 	/* If ret is 1 then we just hit the end of the extent array */
1119 	if (ret == 1)
1120 		ret = 0;
1121 
1122 	inode_unlock(inode);
1123 	return ret;
1124 }
1125