Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * 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 * @for_btnc: inode for B-tree node cache flag
33 * @for_shadow: inode for shadowed page cache flag
34 */
35struct nilfs_iget_args {
36 u64 ino;
37 __u64 cno;
38 struct nilfs_root *root;
39 bool for_gc;
40 bool for_btnc;
41 bool for_shadow;
42};
43
44static int nilfs_iget_test(struct inode *inode, void *opaque);
45
46void nilfs_inode_add_blocks(struct inode *inode, int n)
47{
48 struct nilfs_root *root = NILFS_I(inode)->i_root;
49
50 inode_add_bytes(inode, i_blocksize(inode) * n);
51 if (root)
52 atomic64_add(n, &root->blocks_count);
53}
54
55void nilfs_inode_sub_blocks(struct inode *inode, int n)
56{
57 struct nilfs_root *root = NILFS_I(inode)->i_root;
58
59 inode_sub_bytes(inode, i_blocksize(inode) * n);
60 if (root)
61 atomic64_sub(n, &root->blocks_count);
62}
63
64/**
65 * nilfs_get_block() - get a file block on the filesystem (callback function)
66 * @inode: inode struct of the target file
67 * @blkoff: file block number
68 * @bh_result: buffer head to be mapped on
69 * @create: indicate whether allocating the block or not when it has not
70 * been allocated yet.
71 *
72 * This function does not issue actual read request of the specified data
73 * block. It is done by VFS.
74 */
75int nilfs_get_block(struct inode *inode, sector_t blkoff,
76 struct buffer_head *bh_result, int create)
77{
78 struct nilfs_inode_info *ii = NILFS_I(inode);
79 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
80 __u64 blknum = 0;
81 int err = 0, ret;
82 unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
83
84 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
85 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
86 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
87 if (ret >= 0) { /* found */
88 map_bh(bh_result, inode->i_sb, blknum);
89 if (ret > 0)
90 bh_result->b_size = (ret << inode->i_blkbits);
91 goto out;
92 }
93 /* data block was not found */
94 if (ret == -ENOENT && create) {
95 struct nilfs_transaction_info ti;
96
97 bh_result->b_blocknr = 0;
98 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
99 if (unlikely(err))
100 goto out;
101 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
102 (unsigned long)bh_result);
103 if (unlikely(err != 0)) {
104 if (err == -EEXIST) {
105 /*
106 * The get_block() function could be called
107 * from multiple callers for an inode.
108 * However, the page having this block must
109 * be locked in this case.
110 */
111 nilfs_warn(inode->i_sb,
112 "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
113 __func__, inode->i_ino,
114 (unsigned long long)blkoff);
115 err = 0;
116 }
117 nilfs_transaction_abort(inode->i_sb);
118 goto out;
119 }
120 nilfs_mark_inode_dirty_sync(inode);
121 nilfs_transaction_commit(inode->i_sb); /* never fails */
122 /* Error handling should be detailed */
123 set_buffer_new(bh_result);
124 set_buffer_delay(bh_result);
125 map_bh(bh_result, inode->i_sb, 0);
126 /* Disk block number must be changed to proper value */
127
128 } else if (ret == -ENOENT) {
129 /*
130 * not found is not error (e.g. hole); must return without
131 * the mapped state flag.
132 */
133 ;
134 } else {
135 err = ret;
136 }
137
138 out:
139 return err;
140}
141
142/**
143 * nilfs_read_folio() - implement read_folio() method of nilfs_aops {}
144 * address_space_operations.
145 * @file: file struct of the file to be read
146 * @folio: the folio to be read
147 */
148static int nilfs_read_folio(struct file *file, struct folio *folio)
149{
150 return mpage_read_folio(folio, nilfs_get_block);
151}
152
153static void nilfs_readahead(struct readahead_control *rac)
154{
155 mpage_readahead(rac, nilfs_get_block);
156}
157
158static int nilfs_writepages(struct address_space *mapping,
159 struct writeback_control *wbc)
160{
161 struct inode *inode = mapping->host;
162 int err = 0;
163
164 if (sb_rdonly(inode->i_sb)) {
165 nilfs_clear_dirty_pages(mapping, false);
166 return -EROFS;
167 }
168
169 if (wbc->sync_mode == WB_SYNC_ALL)
170 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
171 wbc->range_start,
172 wbc->range_end);
173 return err;
174}
175
176static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
177{
178 struct inode *inode = page->mapping->host;
179 int err;
180
181 if (sb_rdonly(inode->i_sb)) {
182 /*
183 * It means that filesystem was remounted in read-only
184 * mode because of error or metadata corruption. But we
185 * have dirty pages that try to be flushed in background.
186 * So, here we simply discard this dirty page.
187 */
188 nilfs_clear_dirty_page(page, false);
189 unlock_page(page);
190 return -EROFS;
191 }
192
193 redirty_page_for_writepage(wbc, page);
194 unlock_page(page);
195
196 if (wbc->sync_mode == WB_SYNC_ALL) {
197 err = nilfs_construct_segment(inode->i_sb);
198 if (unlikely(err))
199 return err;
200 } else if (wbc->for_reclaim)
201 nilfs_flush_segment(inode->i_sb, inode->i_ino);
202
203 return 0;
204}
205
206static bool nilfs_dirty_folio(struct address_space *mapping,
207 struct folio *folio)
208{
209 struct inode *inode = mapping->host;
210 struct buffer_head *head;
211 unsigned int nr_dirty = 0;
212 bool ret = filemap_dirty_folio(mapping, folio);
213
214 /*
215 * The page may not be locked, eg if called from try_to_unmap_one()
216 */
217 spin_lock(&mapping->private_lock);
218 head = folio_buffers(folio);
219 if (head) {
220 struct buffer_head *bh = head;
221
222 do {
223 /* Do not mark hole blocks dirty */
224 if (buffer_dirty(bh) || !buffer_mapped(bh))
225 continue;
226
227 set_buffer_dirty(bh);
228 nr_dirty++;
229 } while (bh = bh->b_this_page, bh != head);
230 } else if (ret) {
231 nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
232 }
233 spin_unlock(&mapping->private_lock);
234
235 if (nr_dirty)
236 nilfs_set_file_dirty(inode, nr_dirty);
237 return ret;
238}
239
240void nilfs_write_failed(struct address_space *mapping, loff_t to)
241{
242 struct inode *inode = mapping->host;
243
244 if (to > inode->i_size) {
245 truncate_pagecache(inode, inode->i_size);
246 nilfs_truncate(inode);
247 }
248}
249
250static int nilfs_write_begin(struct file *file, struct address_space *mapping,
251 loff_t pos, unsigned len,
252 struct page **pagep, void **fsdata)
253
254{
255 struct inode *inode = mapping->host;
256 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
257
258 if (unlikely(err))
259 return err;
260
261 err = block_write_begin(mapping, pos, len, pagep, nilfs_get_block);
262 if (unlikely(err)) {
263 nilfs_write_failed(mapping, pos + len);
264 nilfs_transaction_abort(inode->i_sb);
265 }
266 return err;
267}
268
269static int nilfs_write_end(struct file *file, struct address_space *mapping,
270 loff_t pos, unsigned len, unsigned copied,
271 struct page *page, void *fsdata)
272{
273 struct inode *inode = mapping->host;
274 unsigned int start = pos & (PAGE_SIZE - 1);
275 unsigned int nr_dirty;
276 int err;
277
278 nr_dirty = nilfs_page_count_clean_buffers(page, start,
279 start + copied);
280 copied = generic_write_end(file, mapping, pos, len, copied, page,
281 fsdata);
282 nilfs_set_file_dirty(inode, nr_dirty);
283 err = nilfs_transaction_commit(inode->i_sb);
284 return err ? : copied;
285}
286
287static ssize_t
288nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
289{
290 struct inode *inode = file_inode(iocb->ki_filp);
291
292 if (iov_iter_rw(iter) == WRITE)
293 return 0;
294
295 /* Needs synchronization with the cleaner */
296 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
297}
298
299const struct address_space_operations nilfs_aops = {
300 .writepage = nilfs_writepage,
301 .read_folio = nilfs_read_folio,
302 .writepages = nilfs_writepages,
303 .dirty_folio = nilfs_dirty_folio,
304 .readahead = nilfs_readahead,
305 .write_begin = nilfs_write_begin,
306 .write_end = nilfs_write_end,
307 .invalidate_folio = block_invalidate_folio,
308 .direct_IO = nilfs_direct_IO,
309 .is_partially_uptodate = block_is_partially_uptodate,
310};
311
312static 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 = false,
318 .for_btnc = false, .for_shadow = false
319 };
320
321 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
322}
323
324struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
325{
326 struct super_block *sb = dir->i_sb;
327 struct the_nilfs *nilfs = sb->s_fs_info;
328 struct inode *inode;
329 struct nilfs_inode_info *ii;
330 struct nilfs_root *root;
331 struct buffer_head *bh;
332 int err = -ENOMEM;
333 ino_t ino;
334
335 inode = new_inode(sb);
336 if (unlikely(!inode))
337 goto failed;
338
339 mapping_set_gfp_mask(inode->i_mapping,
340 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
341
342 root = NILFS_I(dir)->i_root;
343 ii = NILFS_I(inode);
344 ii->i_state = BIT(NILFS_I_NEW);
345 ii->i_root = root;
346
347 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
348 if (unlikely(err))
349 goto failed_ifile_create_inode;
350 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
351
352 if (unlikely(ino < NILFS_USER_INO)) {
353 nilfs_warn(sb,
354 "inode bitmap is inconsistent for reserved inodes");
355 do {
356 brelse(bh);
357 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
358 if (unlikely(err))
359 goto failed_ifile_create_inode;
360 } while (ino < NILFS_USER_INO);
361
362 nilfs_info(sb, "repaired inode bitmap for reserved inodes");
363 }
364 ii->i_bh = bh;
365
366 atomic64_inc(&root->inodes_count);
367 inode_init_owner(&init_user_ns, inode, dir, mode);
368 inode->i_ino = ino;
369 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
370
371 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
372 err = nilfs_bmap_read(ii->i_bmap, NULL);
373 if (err < 0)
374 goto failed_after_creation;
375
376 set_bit(NILFS_I_BMAP, &ii->i_state);
377 /* No lock is needed; iget() ensures it. */
378 }
379
380 ii->i_flags = nilfs_mask_flags(
381 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
382
383 /* ii->i_file_acl = 0; */
384 /* ii->i_dir_acl = 0; */
385 ii->i_dir_start_lookup = 0;
386 nilfs_set_inode_flags(inode);
387 spin_lock(&nilfs->ns_next_gen_lock);
388 inode->i_generation = nilfs->ns_next_generation++;
389 spin_unlock(&nilfs->ns_next_gen_lock);
390 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
391 err = -EIO;
392 goto failed_after_creation;
393 }
394
395 err = nilfs_init_acl(inode, dir);
396 if (unlikely(err))
397 /*
398 * Never occur. When supporting nilfs_init_acl(),
399 * proper cancellation of above jobs should be considered.
400 */
401 goto failed_after_creation;
402
403 return inode;
404
405 failed_after_creation:
406 clear_nlink(inode);
407 if (inode->i_state & I_NEW)
408 unlock_new_inode(inode);
409 iput(inode); /*
410 * raw_inode will be deleted through
411 * nilfs_evict_inode().
412 */
413 goto failed;
414
415 failed_ifile_create_inode:
416 make_bad_inode(inode);
417 iput(inode);
418 failed:
419 return ERR_PTR(err);
420}
421
422void nilfs_set_inode_flags(struct inode *inode)
423{
424 unsigned int flags = NILFS_I(inode)->i_flags;
425 unsigned int new_fl = 0;
426
427 if (flags & FS_SYNC_FL)
428 new_fl |= S_SYNC;
429 if (flags & FS_APPEND_FL)
430 new_fl |= S_APPEND;
431 if (flags & FS_IMMUTABLE_FL)
432 new_fl |= S_IMMUTABLE;
433 if (flags & FS_NOATIME_FL)
434 new_fl |= S_NOATIME;
435 if (flags & FS_DIRSYNC_FL)
436 new_fl |= S_DIRSYNC;
437 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
438 S_NOATIME | S_DIRSYNC);
439}
440
441int nilfs_read_inode_common(struct inode *inode,
442 struct nilfs_inode *raw_inode)
443{
444 struct nilfs_inode_info *ii = NILFS_I(inode);
445 int err;
446
447 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
448 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
449 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
450 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
451 inode->i_size = le64_to_cpu(raw_inode->i_size);
452 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
453 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
454 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
455 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
456 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
457 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
458 if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
459 return -EIO; /* this inode is for metadata and corrupted */
460 if (inode->i_nlink == 0)
461 return -ESTALE; /* this inode is deleted */
462
463 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
464 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
465#if 0
466 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
467 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
468 0 : le32_to_cpu(raw_inode->i_dir_acl);
469#endif
470 ii->i_dir_start_lookup = 0;
471 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
472
473 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
474 S_ISLNK(inode->i_mode)) {
475 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
476 if (err < 0)
477 return err;
478 set_bit(NILFS_I_BMAP, &ii->i_state);
479 /* No lock is needed; iget() ensures it. */
480 }
481 return 0;
482}
483
484static int __nilfs_read_inode(struct super_block *sb,
485 struct nilfs_root *root, unsigned long ino,
486 struct inode *inode)
487{
488 struct the_nilfs *nilfs = sb->s_fs_info;
489 struct buffer_head *bh;
490 struct nilfs_inode *raw_inode;
491 int err;
492
493 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
494 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
495 if (unlikely(err))
496 goto bad_inode;
497
498 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
499
500 err = nilfs_read_inode_common(inode, raw_inode);
501 if (err)
502 goto failed_unmap;
503
504 if (S_ISREG(inode->i_mode)) {
505 inode->i_op = &nilfs_file_inode_operations;
506 inode->i_fop = &nilfs_file_operations;
507 inode->i_mapping->a_ops = &nilfs_aops;
508 } else if (S_ISDIR(inode->i_mode)) {
509 inode->i_op = &nilfs_dir_inode_operations;
510 inode->i_fop = &nilfs_dir_operations;
511 inode->i_mapping->a_ops = &nilfs_aops;
512 } else if (S_ISLNK(inode->i_mode)) {
513 inode->i_op = &nilfs_symlink_inode_operations;
514 inode_nohighmem(inode);
515 inode->i_mapping->a_ops = &nilfs_aops;
516 } else {
517 inode->i_op = &nilfs_special_inode_operations;
518 init_special_inode(
519 inode, inode->i_mode,
520 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
521 }
522 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
523 brelse(bh);
524 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
525 nilfs_set_inode_flags(inode);
526 mapping_set_gfp_mask(inode->i_mapping,
527 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
528 return 0;
529
530 failed_unmap:
531 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
532 brelse(bh);
533
534 bad_inode:
535 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
536 return err;
537}
538
539static int nilfs_iget_test(struct inode *inode, void *opaque)
540{
541 struct nilfs_iget_args *args = opaque;
542 struct nilfs_inode_info *ii;
543
544 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
545 return 0;
546
547 ii = NILFS_I(inode);
548 if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
549 if (!args->for_btnc)
550 return 0;
551 } else if (args->for_btnc) {
552 return 0;
553 }
554 if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
555 if (!args->for_shadow)
556 return 0;
557 } else if (args->for_shadow) {
558 return 0;
559 }
560
561 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
562 return !args->for_gc;
563
564 return args->for_gc && args->cno == ii->i_cno;
565}
566
567static int nilfs_iget_set(struct inode *inode, void *opaque)
568{
569 struct nilfs_iget_args *args = opaque;
570
571 inode->i_ino = args->ino;
572 NILFS_I(inode)->i_cno = args->cno;
573 NILFS_I(inode)->i_root = args->root;
574 if (args->root && args->ino == NILFS_ROOT_INO)
575 nilfs_get_root(args->root);
576
577 if (args->for_gc)
578 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
579 if (args->for_btnc)
580 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
581 if (args->for_shadow)
582 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
583 return 0;
584}
585
586struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
587 unsigned long ino)
588{
589 struct nilfs_iget_args args = {
590 .ino = ino, .root = root, .cno = 0, .for_gc = false,
591 .for_btnc = false, .for_shadow = false
592 };
593
594 return ilookup5(sb, ino, nilfs_iget_test, &args);
595}
596
597struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
598 unsigned long ino)
599{
600 struct nilfs_iget_args args = {
601 .ino = ino, .root = root, .cno = 0, .for_gc = false,
602 .for_btnc = false, .for_shadow = false
603 };
604
605 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
606}
607
608struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
609 unsigned long ino)
610{
611 struct inode *inode;
612 int err;
613
614 inode = nilfs_iget_locked(sb, root, ino);
615 if (unlikely(!inode))
616 return ERR_PTR(-ENOMEM);
617 if (!(inode->i_state & I_NEW))
618 return inode;
619
620 err = __nilfs_read_inode(sb, root, ino, inode);
621 if (unlikely(err)) {
622 iget_failed(inode);
623 return ERR_PTR(err);
624 }
625 unlock_new_inode(inode);
626 return inode;
627}
628
629struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
630 __u64 cno)
631{
632 struct nilfs_iget_args args = {
633 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
634 .for_btnc = false, .for_shadow = false
635 };
636 struct inode *inode;
637 int err;
638
639 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
640 if (unlikely(!inode))
641 return ERR_PTR(-ENOMEM);
642 if (!(inode->i_state & I_NEW))
643 return inode;
644
645 err = nilfs_init_gcinode(inode);
646 if (unlikely(err)) {
647 iget_failed(inode);
648 return ERR_PTR(err);
649 }
650 unlock_new_inode(inode);
651 return inode;
652}
653
654/**
655 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
656 * @inode: inode object
657 *
658 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
659 * or does nothing if the inode already has it. This function allocates
660 * an additional inode to maintain page cache of B-tree nodes one-on-one.
661 *
662 * Return Value: On success, 0 is returned. On errors, one of the following
663 * negative error code is returned.
664 *
665 * %-ENOMEM - Insufficient memory available.
666 */
667int nilfs_attach_btree_node_cache(struct inode *inode)
668{
669 struct nilfs_inode_info *ii = NILFS_I(inode);
670 struct inode *btnc_inode;
671 struct nilfs_iget_args args;
672
673 if (ii->i_assoc_inode)
674 return 0;
675
676 args.ino = inode->i_ino;
677 args.root = ii->i_root;
678 args.cno = ii->i_cno;
679 args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
680 args.for_btnc = true;
681 args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
682
683 btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
684 nilfs_iget_set, &args);
685 if (unlikely(!btnc_inode))
686 return -ENOMEM;
687 if (btnc_inode->i_state & I_NEW) {
688 nilfs_init_btnc_inode(btnc_inode);
689 unlock_new_inode(btnc_inode);
690 }
691 NILFS_I(btnc_inode)->i_assoc_inode = inode;
692 NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
693 ii->i_assoc_inode = btnc_inode;
694
695 return 0;
696}
697
698/**
699 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
700 * @inode: inode object
701 *
702 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
703 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
704 */
705void nilfs_detach_btree_node_cache(struct inode *inode)
706{
707 struct nilfs_inode_info *ii = NILFS_I(inode);
708 struct inode *btnc_inode = ii->i_assoc_inode;
709
710 if (btnc_inode) {
711 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
712 ii->i_assoc_inode = NULL;
713 iput(btnc_inode);
714 }
715}
716
717/**
718 * nilfs_iget_for_shadow - obtain inode for shadow mapping
719 * @inode: inode object that uses shadow mapping
720 *
721 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
722 * caches for shadow mapping. The page cache for data pages is set up
723 * in one inode and the one for b-tree node pages is set up in the
724 * other inode, which is attached to the former inode.
725 *
726 * Return Value: On success, a pointer to the inode for data pages is
727 * returned. On errors, one of the following negative error code is returned
728 * in a pointer type.
729 *
730 * %-ENOMEM - Insufficient memory available.
731 */
732struct inode *nilfs_iget_for_shadow(struct inode *inode)
733{
734 struct nilfs_iget_args args = {
735 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
736 .for_btnc = false, .for_shadow = true
737 };
738 struct inode *s_inode;
739 int err;
740
741 s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
742 nilfs_iget_set, &args);
743 if (unlikely(!s_inode))
744 return ERR_PTR(-ENOMEM);
745 if (!(s_inode->i_state & I_NEW))
746 return inode;
747
748 NILFS_I(s_inode)->i_flags = 0;
749 memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
750 mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
751
752 err = nilfs_attach_btree_node_cache(s_inode);
753 if (unlikely(err)) {
754 iget_failed(s_inode);
755 return ERR_PTR(err);
756 }
757 unlock_new_inode(s_inode);
758 return s_inode;
759}
760
761void nilfs_write_inode_common(struct inode *inode,
762 struct nilfs_inode *raw_inode, int has_bmap)
763{
764 struct nilfs_inode_info *ii = NILFS_I(inode);
765
766 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
767 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
768 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
769 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
770 raw_inode->i_size = cpu_to_le64(inode->i_size);
771 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
772 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
773 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
774 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
775 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
776
777 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
778 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
779
780 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
781 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
782
783 /* zero-fill unused portion in the case of super root block */
784 raw_inode->i_xattr = 0;
785 raw_inode->i_pad = 0;
786 memset((void *)raw_inode + sizeof(*raw_inode), 0,
787 nilfs->ns_inode_size - sizeof(*raw_inode));
788 }
789
790 if (has_bmap)
791 nilfs_bmap_write(ii->i_bmap, raw_inode);
792 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
793 raw_inode->i_device_code =
794 cpu_to_le64(huge_encode_dev(inode->i_rdev));
795 /*
796 * When extending inode, nilfs->ns_inode_size should be checked
797 * for substitutions of appended fields.
798 */
799}
800
801void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
802{
803 ino_t ino = inode->i_ino;
804 struct nilfs_inode_info *ii = NILFS_I(inode);
805 struct inode *ifile = ii->i_root->ifile;
806 struct nilfs_inode *raw_inode;
807
808 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
809
810 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
811 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
812 if (flags & I_DIRTY_DATASYNC)
813 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
814
815 nilfs_write_inode_common(inode, raw_inode, 0);
816 /*
817 * XXX: call with has_bmap = 0 is a workaround to avoid
818 * deadlock of bmap. This delays update of i_bmap to just
819 * before writing.
820 */
821
822 nilfs_ifile_unmap_inode(ifile, ino, ibh);
823}
824
825#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
826
827static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
828 unsigned long from)
829{
830 __u64 b;
831 int ret;
832
833 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
834 return;
835repeat:
836 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
837 if (ret == -ENOENT)
838 return;
839 else if (ret < 0)
840 goto failed;
841
842 if (b < from)
843 return;
844
845 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
846 ret = nilfs_bmap_truncate(ii->i_bmap, b);
847 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
848 if (!ret || (ret == -ENOMEM &&
849 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
850 goto repeat;
851
852failed:
853 nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
854 ret, ii->vfs_inode.i_ino);
855}
856
857void nilfs_truncate(struct inode *inode)
858{
859 unsigned long blkoff;
860 unsigned int blocksize;
861 struct nilfs_transaction_info ti;
862 struct super_block *sb = inode->i_sb;
863 struct nilfs_inode_info *ii = NILFS_I(inode);
864
865 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
866 return;
867 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
868 return;
869
870 blocksize = sb->s_blocksize;
871 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
872 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
873
874 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
875
876 nilfs_truncate_bmap(ii, blkoff);
877
878 inode->i_mtime = inode->i_ctime = current_time(inode);
879 if (IS_SYNC(inode))
880 nilfs_set_transaction_flag(NILFS_TI_SYNC);
881
882 nilfs_mark_inode_dirty(inode);
883 nilfs_set_file_dirty(inode, 0);
884 nilfs_transaction_commit(sb);
885 /*
886 * May construct a logical segment and may fail in sync mode.
887 * But truncate has no return value.
888 */
889}
890
891static void nilfs_clear_inode(struct inode *inode)
892{
893 struct nilfs_inode_info *ii = NILFS_I(inode);
894
895 /*
896 * Free resources allocated in nilfs_read_inode(), here.
897 */
898 BUG_ON(!list_empty(&ii->i_dirty));
899 brelse(ii->i_bh);
900 ii->i_bh = NULL;
901
902 if (nilfs_is_metadata_file_inode(inode))
903 nilfs_mdt_clear(inode);
904
905 if (test_bit(NILFS_I_BMAP, &ii->i_state))
906 nilfs_bmap_clear(ii->i_bmap);
907
908 if (!test_bit(NILFS_I_BTNC, &ii->i_state))
909 nilfs_detach_btree_node_cache(inode);
910
911 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
912 nilfs_put_root(ii->i_root);
913}
914
915void nilfs_evict_inode(struct inode *inode)
916{
917 struct nilfs_transaction_info ti;
918 struct super_block *sb = inode->i_sb;
919 struct nilfs_inode_info *ii = NILFS_I(inode);
920 int ret;
921
922 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
923 truncate_inode_pages_final(&inode->i_data);
924 clear_inode(inode);
925 nilfs_clear_inode(inode);
926 return;
927 }
928 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
929
930 truncate_inode_pages_final(&inode->i_data);
931
932 /* TODO: some of the following operations may fail. */
933 nilfs_truncate_bmap(ii, 0);
934 nilfs_mark_inode_dirty(inode);
935 clear_inode(inode);
936
937 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
938 if (!ret)
939 atomic64_dec(&ii->i_root->inodes_count);
940
941 nilfs_clear_inode(inode);
942
943 if (IS_SYNC(inode))
944 nilfs_set_transaction_flag(NILFS_TI_SYNC);
945 nilfs_transaction_commit(sb);
946 /*
947 * May construct a logical segment and may fail in sync mode.
948 * But delete_inode has no return value.
949 */
950}
951
952int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
953 struct iattr *iattr)
954{
955 struct nilfs_transaction_info ti;
956 struct inode *inode = d_inode(dentry);
957 struct super_block *sb = inode->i_sb;
958 int err;
959
960 err = setattr_prepare(&init_user_ns, dentry, iattr);
961 if (err)
962 return err;
963
964 err = nilfs_transaction_begin(sb, &ti, 0);
965 if (unlikely(err))
966 return err;
967
968 if ((iattr->ia_valid & ATTR_SIZE) &&
969 iattr->ia_size != i_size_read(inode)) {
970 inode_dio_wait(inode);
971 truncate_setsize(inode, iattr->ia_size);
972 nilfs_truncate(inode);
973 }
974
975 setattr_copy(&init_user_ns, inode, iattr);
976 mark_inode_dirty(inode);
977
978 if (iattr->ia_valid & ATTR_MODE) {
979 err = nilfs_acl_chmod(inode);
980 if (unlikely(err))
981 goto out_err;
982 }
983
984 return nilfs_transaction_commit(sb);
985
986out_err:
987 nilfs_transaction_abort(sb);
988 return err;
989}
990
991int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
992 int mask)
993{
994 struct nilfs_root *root = NILFS_I(inode)->i_root;
995
996 if ((mask & MAY_WRITE) && root &&
997 root->cno != NILFS_CPTREE_CURRENT_CNO)
998 return -EROFS; /* snapshot is not writable */
999
1000 return generic_permission(&init_user_ns, inode, mask);
1001}
1002
1003int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1004{
1005 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1006 struct nilfs_inode_info *ii = NILFS_I(inode);
1007 int err;
1008
1009 spin_lock(&nilfs->ns_inode_lock);
1010 if (ii->i_bh == NULL) {
1011 spin_unlock(&nilfs->ns_inode_lock);
1012 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1013 inode->i_ino, pbh);
1014 if (unlikely(err))
1015 return err;
1016 spin_lock(&nilfs->ns_inode_lock);
1017 if (ii->i_bh == NULL)
1018 ii->i_bh = *pbh;
1019 else {
1020 brelse(*pbh);
1021 *pbh = ii->i_bh;
1022 }
1023 } else
1024 *pbh = ii->i_bh;
1025
1026 get_bh(*pbh);
1027 spin_unlock(&nilfs->ns_inode_lock);
1028 return 0;
1029}
1030
1031int nilfs_inode_dirty(struct inode *inode)
1032{
1033 struct nilfs_inode_info *ii = NILFS_I(inode);
1034 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1035 int ret = 0;
1036
1037 if (!list_empty(&ii->i_dirty)) {
1038 spin_lock(&nilfs->ns_inode_lock);
1039 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1040 test_bit(NILFS_I_BUSY, &ii->i_state);
1041 spin_unlock(&nilfs->ns_inode_lock);
1042 }
1043 return ret;
1044}
1045
1046int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1047{
1048 struct nilfs_inode_info *ii = NILFS_I(inode);
1049 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1050
1051 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1052
1053 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1054 return 0;
1055
1056 spin_lock(&nilfs->ns_inode_lock);
1057 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1058 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1059 /*
1060 * Because this routine may race with nilfs_dispose_list(),
1061 * we have to check NILFS_I_QUEUED here, too.
1062 */
1063 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1064 /*
1065 * This will happen when somebody is freeing
1066 * this inode.
1067 */
1068 nilfs_warn(inode->i_sb,
1069 "cannot set file dirty (ino=%lu): the file is being freed",
1070 inode->i_ino);
1071 spin_unlock(&nilfs->ns_inode_lock);
1072 return -EINVAL; /*
1073 * NILFS_I_DIRTY may remain for
1074 * freeing inode.
1075 */
1076 }
1077 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1078 set_bit(NILFS_I_QUEUED, &ii->i_state);
1079 }
1080 spin_unlock(&nilfs->ns_inode_lock);
1081 return 0;
1082}
1083
1084int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1085{
1086 struct buffer_head *ibh;
1087 int err;
1088
1089 err = nilfs_load_inode_block(inode, &ibh);
1090 if (unlikely(err)) {
1091 nilfs_warn(inode->i_sb,
1092 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1093 inode->i_ino, err);
1094 return err;
1095 }
1096 nilfs_update_inode(inode, ibh, flags);
1097 mark_buffer_dirty(ibh);
1098 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1099 brelse(ibh);
1100 return 0;
1101}
1102
1103/**
1104 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1105 * @inode: inode of the file to be registered.
1106 * @flags: flags to determine the dirty state of the inode
1107 *
1108 * nilfs_dirty_inode() loads a inode block containing the specified
1109 * @inode and copies data from a nilfs_inode to a corresponding inode
1110 * entry in the inode block. This operation is excluded from the segment
1111 * construction. This function can be called both as a single operation
1112 * and as a part of indivisible file operations.
1113 */
1114void nilfs_dirty_inode(struct inode *inode, int flags)
1115{
1116 struct nilfs_transaction_info ti;
1117 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1118
1119 if (is_bad_inode(inode)) {
1120 nilfs_warn(inode->i_sb,
1121 "tried to mark bad_inode dirty. ignored.");
1122 dump_stack();
1123 return;
1124 }
1125 if (mdi) {
1126 nilfs_mdt_mark_dirty(inode);
1127 return;
1128 }
1129 nilfs_transaction_begin(inode->i_sb, &ti, 0);
1130 __nilfs_mark_inode_dirty(inode, flags);
1131 nilfs_transaction_commit(inode->i_sb); /* never fails */
1132}
1133
1134int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1135 __u64 start, __u64 len)
1136{
1137 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1138 __u64 logical = 0, phys = 0, size = 0;
1139 __u32 flags = 0;
1140 loff_t isize;
1141 sector_t blkoff, end_blkoff;
1142 sector_t delalloc_blkoff;
1143 unsigned long delalloc_blklen;
1144 unsigned int blkbits = inode->i_blkbits;
1145 int ret, n;
1146
1147 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1148 if (ret)
1149 return ret;
1150
1151 inode_lock(inode);
1152
1153 isize = i_size_read(inode);
1154
1155 blkoff = start >> blkbits;
1156 end_blkoff = (start + len - 1) >> blkbits;
1157
1158 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1159 &delalloc_blkoff);
1160
1161 do {
1162 __u64 blkphy;
1163 unsigned int maxblocks;
1164
1165 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1166 if (size) {
1167 /* End of the current extent */
1168 ret = fiemap_fill_next_extent(
1169 fieinfo, logical, phys, size, flags);
1170 if (ret)
1171 break;
1172 }
1173 if (blkoff > end_blkoff)
1174 break;
1175
1176 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1177 logical = blkoff << blkbits;
1178 phys = 0;
1179 size = delalloc_blklen << blkbits;
1180
1181 blkoff = delalloc_blkoff + delalloc_blklen;
1182 delalloc_blklen = nilfs_find_uncommitted_extent(
1183 inode, blkoff, &delalloc_blkoff);
1184 continue;
1185 }
1186
1187 /*
1188 * Limit the number of blocks that we look up so as
1189 * not to get into the next delayed allocation extent.
1190 */
1191 maxblocks = INT_MAX;
1192 if (delalloc_blklen)
1193 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1194 maxblocks);
1195 blkphy = 0;
1196
1197 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1198 n = nilfs_bmap_lookup_contig(
1199 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1200 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1201
1202 if (n < 0) {
1203 int past_eof;
1204
1205 if (unlikely(n != -ENOENT))
1206 break; /* error */
1207
1208 /* HOLE */
1209 blkoff++;
1210 past_eof = ((blkoff << blkbits) >= isize);
1211
1212 if (size) {
1213 /* End of the current extent */
1214
1215 if (past_eof)
1216 flags |= FIEMAP_EXTENT_LAST;
1217
1218 ret = fiemap_fill_next_extent(
1219 fieinfo, logical, phys, size, flags);
1220 if (ret)
1221 break;
1222 size = 0;
1223 }
1224 if (blkoff > end_blkoff || past_eof)
1225 break;
1226 } else {
1227 if (size) {
1228 if (phys && blkphy << blkbits == phys + size) {
1229 /* The current extent goes on */
1230 size += n << blkbits;
1231 } else {
1232 /* Terminate the current extent */
1233 ret = fiemap_fill_next_extent(
1234 fieinfo, logical, phys, size,
1235 flags);
1236 if (ret || blkoff > end_blkoff)
1237 break;
1238
1239 /* Start another extent */
1240 flags = FIEMAP_EXTENT_MERGED;
1241 logical = blkoff << blkbits;
1242 phys = blkphy << blkbits;
1243 size = n << blkbits;
1244 }
1245 } else {
1246 /* Start a new extent */
1247 flags = FIEMAP_EXTENT_MERGED;
1248 logical = blkoff << blkbits;
1249 phys = blkphy << blkbits;
1250 size = n << blkbits;
1251 }
1252 blkoff += n;
1253 }
1254 cond_resched();
1255 } while (true);
1256
1257 /* If ret is 1 then we just hit the end of the extent array */
1258 if (ret == 1)
1259 ret = 0;
1260
1261 inode_unlock(inode);
1262 return ret;
1263}
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 */
33struct nilfs_iget_args {
34 u64 ino;
35 __u64 cno;
36 struct nilfs_root *root;
37 int for_gc;
38};
39
40static int nilfs_iget_test(struct inode *inode, void *opaque);
41
42void 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
51void 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 */
71int 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 */
144static int nilfs_readpage(struct file *file, struct page *page)
145{
146 return mpage_readpage(page, nilfs_get_block);
147}
148
149static void nilfs_readahead(struct readahead_control *rac)
150{
151 mpage_readahead(rac, nilfs_get_block);
152}
153
154static 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
172static 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
202static 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
238void 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
248static 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
268static 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
286static ssize_t
287nilfs_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
298const 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
312static 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
323struct 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
406void 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
425int 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
466static 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
521static 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
536static 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
553struct 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
563struct 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
573struct 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
594struct 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
618void 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
658void 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
684static 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;
692repeat:
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
709failed:
710 nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
711 ret, ii->vfs_inode.i_ino);
712}
713
714void 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
748static 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
771void 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
808int 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
842out_err:
843 nilfs_transaction_abort(sb);
844 return err;
845}
846
847int 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
859int 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
887int 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
902int 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
940int __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 */
969void 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
989int 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}