xref: /linux/fs/nilfs2/inode.c (revision 662fa3d6099374c4615bf64d06895e3573b935b2)
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 <linux/fiemap.h>
18 #include "nilfs.h"
19 #include "btnode.h"
20 #include "segment.h"
21 #include "page.h"
22 #include "mdt.h"
23 #include "cpfile.h"
24 #include "ifile.h"
25 
26 /**
27  * struct nilfs_iget_args - arguments used during comparison between inodes
28  * @ino: inode number
29  * @cno: checkpoint number
30  * @root: pointer on NILFS root object (mounted checkpoint)
31  * @for_gc: inode for GC flag
32  */
33 struct nilfs_iget_args {
34 	u64 ino;
35 	__u64 cno;
36 	struct nilfs_root *root;
37 	int for_gc;
38 };
39 
40 static int nilfs_iget_test(struct inode *inode, void *opaque);
41 
42 void nilfs_inode_add_blocks(struct inode *inode, int n)
43 {
44 	struct nilfs_root *root = NILFS_I(inode)->i_root;
45 
46 	inode_add_bytes(inode, i_blocksize(inode) * n);
47 	if (root)
48 		atomic64_add(n, &root->blocks_count);
49 }
50 
51 void nilfs_inode_sub_blocks(struct inode *inode, int n)
52 {
53 	struct nilfs_root *root = NILFS_I(inode)->i_root;
54 
55 	inode_sub_bytes(inode, i_blocksize(inode) * n);
56 	if (root)
57 		atomic64_sub(n, &root->blocks_count);
58 }
59 
60 /**
61  * nilfs_get_block() - get a file block on the filesystem (callback function)
62  * @inode - inode struct of the target file
63  * @blkoff - file block number
64  * @bh_result - buffer head to be mapped on
65  * @create - indicate whether allocating the block or not when it has not
66  *      been allocated yet.
67  *
68  * This function does not issue actual read request of the specified data
69  * block. It is done by VFS.
70  */
71 int nilfs_get_block(struct inode *inode, sector_t blkoff,
72 		    struct buffer_head *bh_result, int create)
73 {
74 	struct nilfs_inode_info *ii = NILFS_I(inode);
75 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
76 	__u64 blknum = 0;
77 	int err = 0, ret;
78 	unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
79 
80 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
81 	ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
82 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
83 	if (ret >= 0) {	/* found */
84 		map_bh(bh_result, inode->i_sb, blknum);
85 		if (ret > 0)
86 			bh_result->b_size = (ret << inode->i_blkbits);
87 		goto out;
88 	}
89 	/* data block was not found */
90 	if (ret == -ENOENT && create) {
91 		struct nilfs_transaction_info ti;
92 
93 		bh_result->b_blocknr = 0;
94 		err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
95 		if (unlikely(err))
96 			goto out;
97 		err = nilfs_bmap_insert(ii->i_bmap, blkoff,
98 					(unsigned long)bh_result);
99 		if (unlikely(err != 0)) {
100 			if (err == -EEXIST) {
101 				/*
102 				 * The get_block() function could be called
103 				 * from multiple callers for an inode.
104 				 * However, the page having this block must
105 				 * be locked in this case.
106 				 */
107 				nilfs_warn(inode->i_sb,
108 					   "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
109 					   __func__, inode->i_ino,
110 					   (unsigned long long)blkoff);
111 				err = 0;
112 			}
113 			nilfs_transaction_abort(inode->i_sb);
114 			goto out;
115 		}
116 		nilfs_mark_inode_dirty_sync(inode);
117 		nilfs_transaction_commit(inode->i_sb); /* never fails */
118 		/* Error handling should be detailed */
119 		set_buffer_new(bh_result);
120 		set_buffer_delay(bh_result);
121 		map_bh(bh_result, inode->i_sb, 0);
122 		/* Disk block number must be changed to proper value */
123 
124 	} else if (ret == -ENOENT) {
125 		/*
126 		 * not found is not error (e.g. hole); must return without
127 		 * the mapped state flag.
128 		 */
129 		;
130 	} else {
131 		err = ret;
132 	}
133 
134  out:
135 	return err;
136 }
137 
138 /**
139  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
140  * address_space_operations.
141  * @file - file struct of the file to be read
142  * @page - the page to be read
143  */
144 static int nilfs_readpage(struct file *file, struct page *page)
145 {
146 	return mpage_readpage(page, nilfs_get_block);
147 }
148 
149 static void nilfs_readahead(struct readahead_control *rac)
150 {
151 	mpage_readahead(rac, nilfs_get_block);
152 }
153 
154 static int nilfs_writepages(struct address_space *mapping,
155 			    struct writeback_control *wbc)
156 {
157 	struct inode *inode = mapping->host;
158 	int err = 0;
159 
160 	if (sb_rdonly(inode->i_sb)) {
161 		nilfs_clear_dirty_pages(mapping, false);
162 		return -EROFS;
163 	}
164 
165 	if (wbc->sync_mode == WB_SYNC_ALL)
166 		err = nilfs_construct_dsync_segment(inode->i_sb, inode,
167 						    wbc->range_start,
168 						    wbc->range_end);
169 	return err;
170 }
171 
172 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
173 {
174 	struct inode *inode = page->mapping->host;
175 	int err;
176 
177 	if (sb_rdonly(inode->i_sb)) {
178 		/*
179 		 * It means that filesystem was remounted in read-only
180 		 * mode because of error or metadata corruption. But we
181 		 * have dirty pages that try to be flushed in background.
182 		 * So, here we simply discard this dirty page.
183 		 */
184 		nilfs_clear_dirty_page(page, false);
185 		unlock_page(page);
186 		return -EROFS;
187 	}
188 
189 	redirty_page_for_writepage(wbc, page);
190 	unlock_page(page);
191 
192 	if (wbc->sync_mode == WB_SYNC_ALL) {
193 		err = nilfs_construct_segment(inode->i_sb);
194 		if (unlikely(err))
195 			return err;
196 	} else if (wbc->for_reclaim)
197 		nilfs_flush_segment(inode->i_sb, inode->i_ino);
198 
199 	return 0;
200 }
201 
202 static int nilfs_set_page_dirty(struct page *page)
203 {
204 	struct inode *inode = page->mapping->host;
205 	int ret = __set_page_dirty_nobuffers(page);
206 
207 	if (page_has_buffers(page)) {
208 		unsigned int nr_dirty = 0;
209 		struct buffer_head *bh, *head;
210 
211 		/*
212 		 * This page is locked by callers, and no other thread
213 		 * concurrently marks its buffers dirty since they are
214 		 * only dirtied through routines in fs/buffer.c in
215 		 * which call sites of mark_buffer_dirty are protected
216 		 * by page lock.
217 		 */
218 		bh = head = page_buffers(page);
219 		do {
220 			/* Do not mark hole blocks dirty */
221 			if (buffer_dirty(bh) || !buffer_mapped(bh))
222 				continue;
223 
224 			set_buffer_dirty(bh);
225 			nr_dirty++;
226 		} while (bh = bh->b_this_page, bh != head);
227 
228 		if (nr_dirty)
229 			nilfs_set_file_dirty(inode, nr_dirty);
230 	} else if (ret) {
231 		unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
232 
233 		nilfs_set_file_dirty(inode, nr_dirty);
234 	}
235 	return ret;
236 }
237 
238 void nilfs_write_failed(struct address_space *mapping, loff_t to)
239 {
240 	struct inode *inode = mapping->host;
241 
242 	if (to > inode->i_size) {
243 		truncate_pagecache(inode, inode->i_size);
244 		nilfs_truncate(inode);
245 	}
246 }
247 
248 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
249 			     loff_t pos, unsigned len, unsigned flags,
250 			     struct page **pagep, void **fsdata)
251 
252 {
253 	struct inode *inode = mapping->host;
254 	int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
255 
256 	if (unlikely(err))
257 		return err;
258 
259 	err = block_write_begin(mapping, pos, len, flags, pagep,
260 				nilfs_get_block);
261 	if (unlikely(err)) {
262 		nilfs_write_failed(mapping, pos + len);
263 		nilfs_transaction_abort(inode->i_sb);
264 	}
265 	return err;
266 }
267 
268 static int nilfs_write_end(struct file *file, struct address_space *mapping,
269 			   loff_t pos, unsigned len, unsigned copied,
270 			   struct page *page, void *fsdata)
271 {
272 	struct inode *inode = mapping->host;
273 	unsigned int start = pos & (PAGE_SIZE - 1);
274 	unsigned int nr_dirty;
275 	int err;
276 
277 	nr_dirty = nilfs_page_count_clean_buffers(page, start,
278 						  start + copied);
279 	copied = generic_write_end(file, mapping, pos, len, copied, page,
280 				   fsdata);
281 	nilfs_set_file_dirty(inode, nr_dirty);
282 	err = nilfs_transaction_commit(inode->i_sb);
283 	return err ? : copied;
284 }
285 
286 static ssize_t
287 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
288 {
289 	struct inode *inode = file_inode(iocb->ki_filp);
290 
291 	if (iov_iter_rw(iter) == WRITE)
292 		return 0;
293 
294 	/* Needs synchronization with the cleaner */
295 	return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
296 }
297 
298 const struct address_space_operations nilfs_aops = {
299 	.writepage		= nilfs_writepage,
300 	.readpage		= nilfs_readpage,
301 	.writepages		= nilfs_writepages,
302 	.set_page_dirty		= nilfs_set_page_dirty,
303 	.readahead		= nilfs_readahead,
304 	.write_begin		= nilfs_write_begin,
305 	.write_end		= nilfs_write_end,
306 	/* .releasepage		= nilfs_releasepage, */
307 	.invalidatepage		= block_invalidatepage,
308 	.direct_IO		= nilfs_direct_IO,
309 	.is_partially_uptodate  = block_is_partially_uptodate,
310 };
311 
312 static int nilfs_insert_inode_locked(struct inode *inode,
313 				     struct nilfs_root *root,
314 				     unsigned long ino)
315 {
316 	struct nilfs_iget_args args = {
317 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
318 	};
319 
320 	return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
321 }
322 
323 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
324 {
325 	struct super_block *sb = dir->i_sb;
326 	struct the_nilfs *nilfs = sb->s_fs_info;
327 	struct inode *inode;
328 	struct nilfs_inode_info *ii;
329 	struct nilfs_root *root;
330 	int err = -ENOMEM;
331 	ino_t ino;
332 
333 	inode = new_inode(sb);
334 	if (unlikely(!inode))
335 		goto failed;
336 
337 	mapping_set_gfp_mask(inode->i_mapping,
338 			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
339 
340 	root = NILFS_I(dir)->i_root;
341 	ii = NILFS_I(inode);
342 	ii->i_state = BIT(NILFS_I_NEW);
343 	ii->i_root = root;
344 
345 	err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
346 	if (unlikely(err))
347 		goto failed_ifile_create_inode;
348 	/* reference count of i_bh inherits from nilfs_mdt_read_block() */
349 
350 	atomic64_inc(&root->inodes_count);
351 	inode_init_owner(&init_user_ns, inode, dir, mode);
352 	inode->i_ino = ino;
353 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
354 
355 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
356 		err = nilfs_bmap_read(ii->i_bmap, NULL);
357 		if (err < 0)
358 			goto failed_after_creation;
359 
360 		set_bit(NILFS_I_BMAP, &ii->i_state);
361 		/* No lock is needed; iget() ensures it. */
362 	}
363 
364 	ii->i_flags = nilfs_mask_flags(
365 		mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
366 
367 	/* ii->i_file_acl = 0; */
368 	/* ii->i_dir_acl = 0; */
369 	ii->i_dir_start_lookup = 0;
370 	nilfs_set_inode_flags(inode);
371 	spin_lock(&nilfs->ns_next_gen_lock);
372 	inode->i_generation = nilfs->ns_next_generation++;
373 	spin_unlock(&nilfs->ns_next_gen_lock);
374 	if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
375 		err = -EIO;
376 		goto failed_after_creation;
377 	}
378 
379 	err = nilfs_init_acl(inode, dir);
380 	if (unlikely(err))
381 		/*
382 		 * Never occur.  When supporting nilfs_init_acl(),
383 		 * proper cancellation of above jobs should be considered.
384 		 */
385 		goto failed_after_creation;
386 
387 	return inode;
388 
389  failed_after_creation:
390 	clear_nlink(inode);
391 	if (inode->i_state & I_NEW)
392 		unlock_new_inode(inode);
393 	iput(inode);  /*
394 		       * raw_inode will be deleted through
395 		       * nilfs_evict_inode().
396 		       */
397 	goto failed;
398 
399  failed_ifile_create_inode:
400 	make_bad_inode(inode);
401 	iput(inode);
402  failed:
403 	return ERR_PTR(err);
404 }
405 
406 void nilfs_set_inode_flags(struct inode *inode)
407 {
408 	unsigned int flags = NILFS_I(inode)->i_flags;
409 	unsigned int new_fl = 0;
410 
411 	if (flags & FS_SYNC_FL)
412 		new_fl |= S_SYNC;
413 	if (flags & FS_APPEND_FL)
414 		new_fl |= S_APPEND;
415 	if (flags & FS_IMMUTABLE_FL)
416 		new_fl |= S_IMMUTABLE;
417 	if (flags & FS_NOATIME_FL)
418 		new_fl |= S_NOATIME;
419 	if (flags & FS_DIRSYNC_FL)
420 		new_fl |= S_DIRSYNC;
421 	inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
422 			S_NOATIME | S_DIRSYNC);
423 }
424 
425 int nilfs_read_inode_common(struct inode *inode,
426 			    struct nilfs_inode *raw_inode)
427 {
428 	struct nilfs_inode_info *ii = NILFS_I(inode);
429 	int err;
430 
431 	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
432 	i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
433 	i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
434 	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
435 	inode->i_size = le64_to_cpu(raw_inode->i_size);
436 	inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
437 	inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
438 	inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
439 	inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
440 	inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
441 	inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
442 	if (inode->i_nlink == 0)
443 		return -ESTALE; /* this inode is deleted */
444 
445 	inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
446 	ii->i_flags = le32_to_cpu(raw_inode->i_flags);
447 #if 0
448 	ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
449 	ii->i_dir_acl = S_ISREG(inode->i_mode) ?
450 		0 : le32_to_cpu(raw_inode->i_dir_acl);
451 #endif
452 	ii->i_dir_start_lookup = 0;
453 	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
454 
455 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
456 	    S_ISLNK(inode->i_mode)) {
457 		err = nilfs_bmap_read(ii->i_bmap, raw_inode);
458 		if (err < 0)
459 			return err;
460 		set_bit(NILFS_I_BMAP, &ii->i_state);
461 		/* No lock is needed; iget() ensures it. */
462 	}
463 	return 0;
464 }
465 
466 static int __nilfs_read_inode(struct super_block *sb,
467 			      struct nilfs_root *root, unsigned long ino,
468 			      struct inode *inode)
469 {
470 	struct the_nilfs *nilfs = sb->s_fs_info;
471 	struct buffer_head *bh;
472 	struct nilfs_inode *raw_inode;
473 	int err;
474 
475 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
476 	err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
477 	if (unlikely(err))
478 		goto bad_inode;
479 
480 	raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
481 
482 	err = nilfs_read_inode_common(inode, raw_inode);
483 	if (err)
484 		goto failed_unmap;
485 
486 	if (S_ISREG(inode->i_mode)) {
487 		inode->i_op = &nilfs_file_inode_operations;
488 		inode->i_fop = &nilfs_file_operations;
489 		inode->i_mapping->a_ops = &nilfs_aops;
490 	} else if (S_ISDIR(inode->i_mode)) {
491 		inode->i_op = &nilfs_dir_inode_operations;
492 		inode->i_fop = &nilfs_dir_operations;
493 		inode->i_mapping->a_ops = &nilfs_aops;
494 	} else if (S_ISLNK(inode->i_mode)) {
495 		inode->i_op = &nilfs_symlink_inode_operations;
496 		inode_nohighmem(inode);
497 		inode->i_mapping->a_ops = &nilfs_aops;
498 	} else {
499 		inode->i_op = &nilfs_special_inode_operations;
500 		init_special_inode(
501 			inode, inode->i_mode,
502 			huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
503 	}
504 	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
505 	brelse(bh);
506 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
507 	nilfs_set_inode_flags(inode);
508 	mapping_set_gfp_mask(inode->i_mapping,
509 			   mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
510 	return 0;
511 
512  failed_unmap:
513 	nilfs_ifile_unmap_inode(root->ifile, ino, bh);
514 	brelse(bh);
515 
516  bad_inode:
517 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
518 	return err;
519 }
520 
521 static int nilfs_iget_test(struct inode *inode, void *opaque)
522 {
523 	struct nilfs_iget_args *args = opaque;
524 	struct nilfs_inode_info *ii;
525 
526 	if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
527 		return 0;
528 
529 	ii = NILFS_I(inode);
530 	if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
531 		return !args->for_gc;
532 
533 	return args->for_gc && args->cno == ii->i_cno;
534 }
535 
536 static int nilfs_iget_set(struct inode *inode, void *opaque)
537 {
538 	struct nilfs_iget_args *args = opaque;
539 
540 	inode->i_ino = args->ino;
541 	if (args->for_gc) {
542 		NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
543 		NILFS_I(inode)->i_cno = args->cno;
544 		NILFS_I(inode)->i_root = NULL;
545 	} else {
546 		if (args->root && args->ino == NILFS_ROOT_INO)
547 			nilfs_get_root(args->root);
548 		NILFS_I(inode)->i_root = args->root;
549 	}
550 	return 0;
551 }
552 
553 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
554 			    unsigned long ino)
555 {
556 	struct nilfs_iget_args args = {
557 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
558 	};
559 
560 	return ilookup5(sb, ino, nilfs_iget_test, &args);
561 }
562 
563 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
564 				unsigned long ino)
565 {
566 	struct nilfs_iget_args args = {
567 		.ino = ino, .root = root, .cno = 0, .for_gc = 0
568 	};
569 
570 	return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
571 }
572 
573 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
574 			 unsigned long ino)
575 {
576 	struct inode *inode;
577 	int err;
578 
579 	inode = nilfs_iget_locked(sb, root, ino);
580 	if (unlikely(!inode))
581 		return ERR_PTR(-ENOMEM);
582 	if (!(inode->i_state & I_NEW))
583 		return inode;
584 
585 	err = __nilfs_read_inode(sb, root, ino, inode);
586 	if (unlikely(err)) {
587 		iget_failed(inode);
588 		return ERR_PTR(err);
589 	}
590 	unlock_new_inode(inode);
591 	return inode;
592 }
593 
594 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
595 				__u64 cno)
596 {
597 	struct nilfs_iget_args args = {
598 		.ino = ino, .root = NULL, .cno = cno, .for_gc = 1
599 	};
600 	struct inode *inode;
601 	int err;
602 
603 	inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
604 	if (unlikely(!inode))
605 		return ERR_PTR(-ENOMEM);
606 	if (!(inode->i_state & I_NEW))
607 		return inode;
608 
609 	err = nilfs_init_gcinode(inode);
610 	if (unlikely(err)) {
611 		iget_failed(inode);
612 		return ERR_PTR(err);
613 	}
614 	unlock_new_inode(inode);
615 	return inode;
616 }
617 
618 void nilfs_write_inode_common(struct inode *inode,
619 			      struct nilfs_inode *raw_inode, int has_bmap)
620 {
621 	struct nilfs_inode_info *ii = NILFS_I(inode);
622 
623 	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
624 	raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
625 	raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
626 	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
627 	raw_inode->i_size = cpu_to_le64(inode->i_size);
628 	raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
629 	raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
630 	raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
631 	raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
632 	raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
633 
634 	raw_inode->i_flags = cpu_to_le32(ii->i_flags);
635 	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
636 
637 	if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
638 		struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
639 
640 		/* zero-fill unused portion in the case of super root block */
641 		raw_inode->i_xattr = 0;
642 		raw_inode->i_pad = 0;
643 		memset((void *)raw_inode + sizeof(*raw_inode), 0,
644 		       nilfs->ns_inode_size - sizeof(*raw_inode));
645 	}
646 
647 	if (has_bmap)
648 		nilfs_bmap_write(ii->i_bmap, raw_inode);
649 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
650 		raw_inode->i_device_code =
651 			cpu_to_le64(huge_encode_dev(inode->i_rdev));
652 	/*
653 	 * When extending inode, nilfs->ns_inode_size should be checked
654 	 * for substitutions of appended fields.
655 	 */
656 }
657 
658 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
659 {
660 	ino_t ino = inode->i_ino;
661 	struct nilfs_inode_info *ii = NILFS_I(inode);
662 	struct inode *ifile = ii->i_root->ifile;
663 	struct nilfs_inode *raw_inode;
664 
665 	raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
666 
667 	if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
668 		memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
669 	if (flags & I_DIRTY_DATASYNC)
670 		set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
671 
672 	nilfs_write_inode_common(inode, raw_inode, 0);
673 		/*
674 		 * XXX: call with has_bmap = 0 is a workaround to avoid
675 		 * deadlock of bmap.  This delays update of i_bmap to just
676 		 * before writing.
677 		 */
678 
679 	nilfs_ifile_unmap_inode(ifile, ino, ibh);
680 }
681 
682 #define NILFS_MAX_TRUNCATE_BLOCKS	16384  /* 64MB for 4KB block */
683 
684 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
685 				unsigned long from)
686 {
687 	__u64 b;
688 	int ret;
689 
690 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
691 		return;
692 repeat:
693 	ret = nilfs_bmap_last_key(ii->i_bmap, &b);
694 	if (ret == -ENOENT)
695 		return;
696 	else if (ret < 0)
697 		goto failed;
698 
699 	if (b < from)
700 		return;
701 
702 	b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
703 	ret = nilfs_bmap_truncate(ii->i_bmap, b);
704 	nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
705 	if (!ret || (ret == -ENOMEM &&
706 		     nilfs_bmap_truncate(ii->i_bmap, b) == 0))
707 		goto repeat;
708 
709 failed:
710 	nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
711 		   ret, ii->vfs_inode.i_ino);
712 }
713 
714 void nilfs_truncate(struct inode *inode)
715 {
716 	unsigned long blkoff;
717 	unsigned int blocksize;
718 	struct nilfs_transaction_info ti;
719 	struct super_block *sb = inode->i_sb;
720 	struct nilfs_inode_info *ii = NILFS_I(inode);
721 
722 	if (!test_bit(NILFS_I_BMAP, &ii->i_state))
723 		return;
724 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
725 		return;
726 
727 	blocksize = sb->s_blocksize;
728 	blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
729 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
730 
731 	block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
732 
733 	nilfs_truncate_bmap(ii, blkoff);
734 
735 	inode->i_mtime = inode->i_ctime = current_time(inode);
736 	if (IS_SYNC(inode))
737 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
738 
739 	nilfs_mark_inode_dirty(inode);
740 	nilfs_set_file_dirty(inode, 0);
741 	nilfs_transaction_commit(sb);
742 	/*
743 	 * May construct a logical segment and may fail in sync mode.
744 	 * But truncate has no return value.
745 	 */
746 }
747 
748 static void nilfs_clear_inode(struct inode *inode)
749 {
750 	struct nilfs_inode_info *ii = NILFS_I(inode);
751 
752 	/*
753 	 * Free resources allocated in nilfs_read_inode(), here.
754 	 */
755 	BUG_ON(!list_empty(&ii->i_dirty));
756 	brelse(ii->i_bh);
757 	ii->i_bh = NULL;
758 
759 	if (nilfs_is_metadata_file_inode(inode))
760 		nilfs_mdt_clear(inode);
761 
762 	if (test_bit(NILFS_I_BMAP, &ii->i_state))
763 		nilfs_bmap_clear(ii->i_bmap);
764 
765 	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
766 
767 	if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
768 		nilfs_put_root(ii->i_root);
769 }
770 
771 void nilfs_evict_inode(struct inode *inode)
772 {
773 	struct nilfs_transaction_info ti;
774 	struct super_block *sb = inode->i_sb;
775 	struct nilfs_inode_info *ii = NILFS_I(inode);
776 	int ret;
777 
778 	if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
779 		truncate_inode_pages_final(&inode->i_data);
780 		clear_inode(inode);
781 		nilfs_clear_inode(inode);
782 		return;
783 	}
784 	nilfs_transaction_begin(sb, &ti, 0); /* never fails */
785 
786 	truncate_inode_pages_final(&inode->i_data);
787 
788 	/* TODO: some of the following operations may fail.  */
789 	nilfs_truncate_bmap(ii, 0);
790 	nilfs_mark_inode_dirty(inode);
791 	clear_inode(inode);
792 
793 	ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
794 	if (!ret)
795 		atomic64_dec(&ii->i_root->inodes_count);
796 
797 	nilfs_clear_inode(inode);
798 
799 	if (IS_SYNC(inode))
800 		nilfs_set_transaction_flag(NILFS_TI_SYNC);
801 	nilfs_transaction_commit(sb);
802 	/*
803 	 * May construct a logical segment and may fail in sync mode.
804 	 * But delete_inode has no return value.
805 	 */
806 }
807 
808 int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
809 		  struct iattr *iattr)
810 {
811 	struct nilfs_transaction_info ti;
812 	struct inode *inode = d_inode(dentry);
813 	struct super_block *sb = inode->i_sb;
814 	int err;
815 
816 	err = setattr_prepare(&init_user_ns, dentry, iattr);
817 	if (err)
818 		return err;
819 
820 	err = nilfs_transaction_begin(sb, &ti, 0);
821 	if (unlikely(err))
822 		return err;
823 
824 	if ((iattr->ia_valid & ATTR_SIZE) &&
825 	    iattr->ia_size != i_size_read(inode)) {
826 		inode_dio_wait(inode);
827 		truncate_setsize(inode, iattr->ia_size);
828 		nilfs_truncate(inode);
829 	}
830 
831 	setattr_copy(&init_user_ns, inode, iattr);
832 	mark_inode_dirty(inode);
833 
834 	if (iattr->ia_valid & ATTR_MODE) {
835 		err = nilfs_acl_chmod(inode);
836 		if (unlikely(err))
837 			goto out_err;
838 	}
839 
840 	return nilfs_transaction_commit(sb);
841 
842 out_err:
843 	nilfs_transaction_abort(sb);
844 	return err;
845 }
846 
847 int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
848 		     int mask)
849 {
850 	struct nilfs_root *root = NILFS_I(inode)->i_root;
851 
852 	if ((mask & MAY_WRITE) && root &&
853 	    root->cno != NILFS_CPTREE_CURRENT_CNO)
854 		return -EROFS; /* snapshot is not writable */
855 
856 	return generic_permission(&init_user_ns, inode, mask);
857 }
858 
859 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
860 {
861 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
862 	struct nilfs_inode_info *ii = NILFS_I(inode);
863 	int err;
864 
865 	spin_lock(&nilfs->ns_inode_lock);
866 	if (ii->i_bh == NULL) {
867 		spin_unlock(&nilfs->ns_inode_lock);
868 		err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
869 						  inode->i_ino, pbh);
870 		if (unlikely(err))
871 			return err;
872 		spin_lock(&nilfs->ns_inode_lock);
873 		if (ii->i_bh == NULL)
874 			ii->i_bh = *pbh;
875 		else {
876 			brelse(*pbh);
877 			*pbh = ii->i_bh;
878 		}
879 	} else
880 		*pbh = ii->i_bh;
881 
882 	get_bh(*pbh);
883 	spin_unlock(&nilfs->ns_inode_lock);
884 	return 0;
885 }
886 
887 int nilfs_inode_dirty(struct inode *inode)
888 {
889 	struct nilfs_inode_info *ii = NILFS_I(inode);
890 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
891 	int ret = 0;
892 
893 	if (!list_empty(&ii->i_dirty)) {
894 		spin_lock(&nilfs->ns_inode_lock);
895 		ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
896 			test_bit(NILFS_I_BUSY, &ii->i_state);
897 		spin_unlock(&nilfs->ns_inode_lock);
898 	}
899 	return ret;
900 }
901 
902 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
903 {
904 	struct nilfs_inode_info *ii = NILFS_I(inode);
905 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
906 
907 	atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
908 
909 	if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
910 		return 0;
911 
912 	spin_lock(&nilfs->ns_inode_lock);
913 	if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
914 	    !test_bit(NILFS_I_BUSY, &ii->i_state)) {
915 		/*
916 		 * Because this routine may race with nilfs_dispose_list(),
917 		 * we have to check NILFS_I_QUEUED here, too.
918 		 */
919 		if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
920 			/*
921 			 * This will happen when somebody is freeing
922 			 * this inode.
923 			 */
924 			nilfs_warn(inode->i_sb,
925 				   "cannot set file dirty (ino=%lu): the file is being freed",
926 				   inode->i_ino);
927 			spin_unlock(&nilfs->ns_inode_lock);
928 			return -EINVAL; /*
929 					 * NILFS_I_DIRTY may remain for
930 					 * freeing inode.
931 					 */
932 		}
933 		list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
934 		set_bit(NILFS_I_QUEUED, &ii->i_state);
935 	}
936 	spin_unlock(&nilfs->ns_inode_lock);
937 	return 0;
938 }
939 
940 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
941 {
942 	struct buffer_head *ibh;
943 	int err;
944 
945 	err = nilfs_load_inode_block(inode, &ibh);
946 	if (unlikely(err)) {
947 		nilfs_warn(inode->i_sb,
948 			   "cannot mark inode dirty (ino=%lu): error %d loading inode block",
949 			   inode->i_ino, err);
950 		return err;
951 	}
952 	nilfs_update_inode(inode, ibh, flags);
953 	mark_buffer_dirty(ibh);
954 	nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
955 	brelse(ibh);
956 	return 0;
957 }
958 
959 /**
960  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
961  * @inode: inode of the file to be registered.
962  *
963  * nilfs_dirty_inode() loads a inode block containing the specified
964  * @inode and copies data from a nilfs_inode to a corresponding inode
965  * entry in the inode block. This operation is excluded from the segment
966  * construction. This function can be called both as a single operation
967  * and as a part of indivisible file operations.
968  */
969 void nilfs_dirty_inode(struct inode *inode, int flags)
970 {
971 	struct nilfs_transaction_info ti;
972 	struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
973 
974 	if (is_bad_inode(inode)) {
975 		nilfs_warn(inode->i_sb,
976 			   "tried to mark bad_inode dirty. ignored.");
977 		dump_stack();
978 		return;
979 	}
980 	if (mdi) {
981 		nilfs_mdt_mark_dirty(inode);
982 		return;
983 	}
984 	nilfs_transaction_begin(inode->i_sb, &ti, 0);
985 	__nilfs_mark_inode_dirty(inode, flags);
986 	nilfs_transaction_commit(inode->i_sb); /* never fails */
987 }
988 
989 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
990 		 __u64 start, __u64 len)
991 {
992 	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
993 	__u64 logical = 0, phys = 0, size = 0;
994 	__u32 flags = 0;
995 	loff_t isize;
996 	sector_t blkoff, end_blkoff;
997 	sector_t delalloc_blkoff;
998 	unsigned long delalloc_blklen;
999 	unsigned int blkbits = inode->i_blkbits;
1000 	int ret, n;
1001 
1002 	ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1003 	if (ret)
1004 		return ret;
1005 
1006 	inode_lock(inode);
1007 
1008 	isize = i_size_read(inode);
1009 
1010 	blkoff = start >> blkbits;
1011 	end_blkoff = (start + len - 1) >> blkbits;
1012 
1013 	delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1014 							&delalloc_blkoff);
1015 
1016 	do {
1017 		__u64 blkphy;
1018 		unsigned int maxblocks;
1019 
1020 		if (delalloc_blklen && blkoff == delalloc_blkoff) {
1021 			if (size) {
1022 				/* End of the current extent */
1023 				ret = fiemap_fill_next_extent(
1024 					fieinfo, logical, phys, size, flags);
1025 				if (ret)
1026 					break;
1027 			}
1028 			if (blkoff > end_blkoff)
1029 				break;
1030 
1031 			flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1032 			logical = blkoff << blkbits;
1033 			phys = 0;
1034 			size = delalloc_blklen << blkbits;
1035 
1036 			blkoff = delalloc_blkoff + delalloc_blklen;
1037 			delalloc_blklen = nilfs_find_uncommitted_extent(
1038 				inode, blkoff, &delalloc_blkoff);
1039 			continue;
1040 		}
1041 
1042 		/*
1043 		 * Limit the number of blocks that we look up so as
1044 		 * not to get into the next delayed allocation extent.
1045 		 */
1046 		maxblocks = INT_MAX;
1047 		if (delalloc_blklen)
1048 			maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1049 					  maxblocks);
1050 		blkphy = 0;
1051 
1052 		down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1053 		n = nilfs_bmap_lookup_contig(
1054 			NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1055 		up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1056 
1057 		if (n < 0) {
1058 			int past_eof;
1059 
1060 			if (unlikely(n != -ENOENT))
1061 				break; /* error */
1062 
1063 			/* HOLE */
1064 			blkoff++;
1065 			past_eof = ((blkoff << blkbits) >= isize);
1066 
1067 			if (size) {
1068 				/* End of the current extent */
1069 
1070 				if (past_eof)
1071 					flags |= FIEMAP_EXTENT_LAST;
1072 
1073 				ret = fiemap_fill_next_extent(
1074 					fieinfo, logical, phys, size, flags);
1075 				if (ret)
1076 					break;
1077 				size = 0;
1078 			}
1079 			if (blkoff > end_blkoff || past_eof)
1080 				break;
1081 		} else {
1082 			if (size) {
1083 				if (phys && blkphy << blkbits == phys + size) {
1084 					/* The current extent goes on */
1085 					size += n << blkbits;
1086 				} else {
1087 					/* Terminate the current extent */
1088 					ret = fiemap_fill_next_extent(
1089 						fieinfo, logical, phys, size,
1090 						flags);
1091 					if (ret || blkoff > end_blkoff)
1092 						break;
1093 
1094 					/* Start another extent */
1095 					flags = FIEMAP_EXTENT_MERGED;
1096 					logical = blkoff << blkbits;
1097 					phys = blkphy << blkbits;
1098 					size = n << blkbits;
1099 				}
1100 			} else {
1101 				/* Start a new extent */
1102 				flags = FIEMAP_EXTENT_MERGED;
1103 				logical = blkoff << blkbits;
1104 				phys = blkphy << blkbits;
1105 				size = n << blkbits;
1106 			}
1107 			blkoff += n;
1108 		}
1109 		cond_resched();
1110 	} while (true);
1111 
1112 	/* If ret is 1 then we just hit the end of the extent array */
1113 	if (ret == 1)
1114 		ret = 0;
1115 
1116 	inode_unlock(inode);
1117 	return ret;
1118 }
1119