Loading...
1/*
2 * inode.c - NILFS inode operations.
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * Written by Ryusuke Konishi.
17 *
18 */
19
20#include <linux/buffer_head.h>
21#include <linux/gfp.h>
22#include <linux/mpage.h>
23#include <linux/pagemap.h>
24#include <linux/writeback.h>
25#include <linux/uio.h>
26#include "nilfs.h"
27#include "btnode.h"
28#include "segment.h"
29#include "page.h"
30#include "mdt.h"
31#include "cpfile.h"
32#include "ifile.h"
33
34/**
35 * struct nilfs_iget_args - arguments used during comparison between inodes
36 * @ino: inode number
37 * @cno: checkpoint number
38 * @root: pointer on NILFS root object (mounted checkpoint)
39 * @for_gc: inode for GC flag
40 */
41struct nilfs_iget_args {
42 u64 ino;
43 __u64 cno;
44 struct nilfs_root *root;
45 int for_gc;
46};
47
48static int nilfs_iget_test(struct inode *inode, void *opaque);
49
50void nilfs_inode_add_blocks(struct inode *inode, int n)
51{
52 struct nilfs_root *root = NILFS_I(inode)->i_root;
53
54 inode_add_bytes(inode, i_blocksize(inode) * n);
55 if (root)
56 atomic64_add(n, &root->blocks_count);
57}
58
59void nilfs_inode_sub_blocks(struct inode *inode, int n)
60{
61 struct nilfs_root *root = NILFS_I(inode)->i_root;
62
63 inode_sub_bytes(inode, i_blocksize(inode) * n);
64 if (root)
65 atomic64_sub(n, &root->blocks_count);
66}
67
68/**
69 * nilfs_get_block() - get a file block on the filesystem (callback function)
70 * @inode - inode struct of the target file
71 * @blkoff - file block number
72 * @bh_result - buffer head to be mapped on
73 * @create - indicate whether allocating the block or not when it has not
74 * been allocated yet.
75 *
76 * This function does not issue actual read request of the specified data
77 * block. It is done by VFS.
78 */
79int nilfs_get_block(struct inode *inode, sector_t blkoff,
80 struct buffer_head *bh_result, int create)
81{
82 struct nilfs_inode_info *ii = NILFS_I(inode);
83 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
84 __u64 blknum = 0;
85 int err = 0, ret;
86 unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
87
88 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
89 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
90 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
91 if (ret >= 0) { /* found */
92 map_bh(bh_result, inode->i_sb, blknum);
93 if (ret > 0)
94 bh_result->b_size = (ret << inode->i_blkbits);
95 goto out;
96 }
97 /* data block was not found */
98 if (ret == -ENOENT && create) {
99 struct nilfs_transaction_info ti;
100
101 bh_result->b_blocknr = 0;
102 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
103 if (unlikely(err))
104 goto out;
105 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
106 (unsigned long)bh_result);
107 if (unlikely(err != 0)) {
108 if (err == -EEXIST) {
109 /*
110 * The get_block() function could be called
111 * from multiple callers for an inode.
112 * However, the page having this block must
113 * be locked in this case.
114 */
115 nilfs_msg(inode->i_sb, KERN_WARNING,
116 "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
117 __func__, inode->i_ino,
118 (unsigned long long)blkoff);
119 err = 0;
120 }
121 nilfs_transaction_abort(inode->i_sb);
122 goto out;
123 }
124 nilfs_mark_inode_dirty_sync(inode);
125 nilfs_transaction_commit(inode->i_sb); /* never fails */
126 /* Error handling should be detailed */
127 set_buffer_new(bh_result);
128 set_buffer_delay(bh_result);
129 map_bh(bh_result, inode->i_sb, 0);
130 /* Disk block number must be changed to proper value */
131
132 } else if (ret == -ENOENT) {
133 /*
134 * not found is not error (e.g. hole); must return without
135 * the mapped state flag.
136 */
137 ;
138 } else {
139 err = ret;
140 }
141
142 out:
143 return err;
144}
145
146/**
147 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
148 * address_space_operations.
149 * @file - file struct of the file to be read
150 * @page - the page to be read
151 */
152static int nilfs_readpage(struct file *file, struct page *page)
153{
154 return mpage_readpage(page, nilfs_get_block);
155}
156
157/**
158 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
159 * address_space_operations.
160 * @file - file struct of the file to be read
161 * @mapping - address_space struct used for reading multiple pages
162 * @pages - the pages to be read
163 * @nr_pages - number of pages to be read
164 */
165static int nilfs_readpages(struct file *file, struct address_space *mapping,
166 struct list_head *pages, unsigned int nr_pages)
167{
168 return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
169}
170
171static int nilfs_writepages(struct address_space *mapping,
172 struct writeback_control *wbc)
173{
174 struct inode *inode = mapping->host;
175 int err = 0;
176
177 if (sb_rdonly(inode->i_sb)) {
178 nilfs_clear_dirty_pages(mapping, false);
179 return -EROFS;
180 }
181
182 if (wbc->sync_mode == WB_SYNC_ALL)
183 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
184 wbc->range_start,
185 wbc->range_end);
186 return err;
187}
188
189static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
190{
191 struct inode *inode = page->mapping->host;
192 int err;
193
194 if (sb_rdonly(inode->i_sb)) {
195 /*
196 * It means that filesystem was remounted in read-only
197 * mode because of error or metadata corruption. But we
198 * have dirty pages that try to be flushed in background.
199 * So, here we simply discard this dirty page.
200 */
201 nilfs_clear_dirty_page(page, false);
202 unlock_page(page);
203 return -EROFS;
204 }
205
206 redirty_page_for_writepage(wbc, page);
207 unlock_page(page);
208
209 if (wbc->sync_mode == WB_SYNC_ALL) {
210 err = nilfs_construct_segment(inode->i_sb);
211 if (unlikely(err))
212 return err;
213 } else if (wbc->for_reclaim)
214 nilfs_flush_segment(inode->i_sb, inode->i_ino);
215
216 return 0;
217}
218
219static int nilfs_set_page_dirty(struct page *page)
220{
221 struct inode *inode = page->mapping->host;
222 int ret = __set_page_dirty_nobuffers(page);
223
224 if (page_has_buffers(page)) {
225 unsigned int nr_dirty = 0;
226 struct buffer_head *bh, *head;
227
228 /*
229 * This page is locked by callers, and no other thread
230 * concurrently marks its buffers dirty since they are
231 * only dirtied through routines in fs/buffer.c in
232 * which call sites of mark_buffer_dirty are protected
233 * by page lock.
234 */
235 bh = head = page_buffers(page);
236 do {
237 /* Do not mark hole blocks dirty */
238 if (buffer_dirty(bh) || !buffer_mapped(bh))
239 continue;
240
241 set_buffer_dirty(bh);
242 nr_dirty++;
243 } while (bh = bh->b_this_page, bh != head);
244
245 if (nr_dirty)
246 nilfs_set_file_dirty(inode, nr_dirty);
247 } else if (ret) {
248 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
249
250 nilfs_set_file_dirty(inode, nr_dirty);
251 }
252 return ret;
253}
254
255void nilfs_write_failed(struct address_space *mapping, loff_t to)
256{
257 struct inode *inode = mapping->host;
258
259 if (to > inode->i_size) {
260 truncate_pagecache(inode, inode->i_size);
261 nilfs_truncate(inode);
262 }
263}
264
265static int nilfs_write_begin(struct file *file, struct address_space *mapping,
266 loff_t pos, unsigned len, unsigned flags,
267 struct page **pagep, void **fsdata)
268
269{
270 struct inode *inode = mapping->host;
271 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
272
273 if (unlikely(err))
274 return err;
275
276 err = block_write_begin(mapping, pos, len, flags, pagep,
277 nilfs_get_block);
278 if (unlikely(err)) {
279 nilfs_write_failed(mapping, pos + len);
280 nilfs_transaction_abort(inode->i_sb);
281 }
282 return err;
283}
284
285static int nilfs_write_end(struct file *file, struct address_space *mapping,
286 loff_t pos, unsigned len, unsigned copied,
287 struct page *page, void *fsdata)
288{
289 struct inode *inode = mapping->host;
290 unsigned int start = pos & (PAGE_SIZE - 1);
291 unsigned int nr_dirty;
292 int err;
293
294 nr_dirty = nilfs_page_count_clean_buffers(page, start,
295 start + copied);
296 copied = generic_write_end(file, mapping, pos, len, copied, page,
297 fsdata);
298 nilfs_set_file_dirty(inode, nr_dirty);
299 err = nilfs_transaction_commit(inode->i_sb);
300 return err ? : copied;
301}
302
303static ssize_t
304nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
305{
306 struct inode *inode = file_inode(iocb->ki_filp);
307
308 if (iov_iter_rw(iter) == WRITE)
309 return 0;
310
311 /* Needs synchronization with the cleaner */
312 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
313}
314
315const struct address_space_operations nilfs_aops = {
316 .writepage = nilfs_writepage,
317 .readpage = nilfs_readpage,
318 .writepages = nilfs_writepages,
319 .set_page_dirty = nilfs_set_page_dirty,
320 .readpages = nilfs_readpages,
321 .write_begin = nilfs_write_begin,
322 .write_end = nilfs_write_end,
323 /* .releasepage = nilfs_releasepage, */
324 .invalidatepage = block_invalidatepage,
325 .direct_IO = nilfs_direct_IO,
326 .is_partially_uptodate = block_is_partially_uptodate,
327};
328
329static int nilfs_insert_inode_locked(struct inode *inode,
330 struct nilfs_root *root,
331 unsigned long ino)
332{
333 struct nilfs_iget_args args = {
334 .ino = ino, .root = root, .cno = 0, .for_gc = 0
335 };
336
337 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
338}
339
340struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
341{
342 struct super_block *sb = dir->i_sb;
343 struct the_nilfs *nilfs = sb->s_fs_info;
344 struct inode *inode;
345 struct nilfs_inode_info *ii;
346 struct nilfs_root *root;
347 int err = -ENOMEM;
348 ino_t ino;
349
350 inode = new_inode(sb);
351 if (unlikely(!inode))
352 goto failed;
353
354 mapping_set_gfp_mask(inode->i_mapping,
355 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
356
357 root = NILFS_I(dir)->i_root;
358 ii = NILFS_I(inode);
359 ii->i_state = BIT(NILFS_I_NEW);
360 ii->i_root = root;
361
362 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
363 if (unlikely(err))
364 goto failed_ifile_create_inode;
365 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
366
367 atomic64_inc(&root->inodes_count);
368 inode_init_owner(inode, dir, mode);
369 inode->i_ino = ino;
370 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
371
372 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
373 err = nilfs_bmap_read(ii->i_bmap, NULL);
374 if (err < 0)
375 goto failed_after_creation;
376
377 set_bit(NILFS_I_BMAP, &ii->i_state);
378 /* No lock is needed; iget() ensures it. */
379 }
380
381 ii->i_flags = nilfs_mask_flags(
382 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
383
384 /* ii->i_file_acl = 0; */
385 /* ii->i_dir_acl = 0; */
386 ii->i_dir_start_lookup = 0;
387 nilfs_set_inode_flags(inode);
388 spin_lock(&nilfs->ns_next_gen_lock);
389 inode->i_generation = nilfs->ns_next_generation++;
390 spin_unlock(&nilfs->ns_next_gen_lock);
391 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
392 err = -EIO;
393 goto failed_after_creation;
394 }
395
396 err = nilfs_init_acl(inode, dir);
397 if (unlikely(err))
398 /*
399 * Never occur. When supporting nilfs_init_acl(),
400 * proper cancellation of above jobs should be considered.
401 */
402 goto failed_after_creation;
403
404 return inode;
405
406 failed_after_creation:
407 clear_nlink(inode);
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 (inode->i_nlink == 0)
459 return -ESTALE; /* this inode is deleted */
460
461 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
462 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
463#if 0
464 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
465 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
466 0 : le32_to_cpu(raw_inode->i_dir_acl);
467#endif
468 ii->i_dir_start_lookup = 0;
469 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
470
471 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
472 S_ISLNK(inode->i_mode)) {
473 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
474 if (err < 0)
475 return err;
476 set_bit(NILFS_I_BMAP, &ii->i_state);
477 /* No lock is needed; iget() ensures it. */
478 }
479 return 0;
480}
481
482static int __nilfs_read_inode(struct super_block *sb,
483 struct nilfs_root *root, unsigned long ino,
484 struct inode *inode)
485{
486 struct the_nilfs *nilfs = sb->s_fs_info;
487 struct buffer_head *bh;
488 struct nilfs_inode *raw_inode;
489 int err;
490
491 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
492 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
493 if (unlikely(err))
494 goto bad_inode;
495
496 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
497
498 err = nilfs_read_inode_common(inode, raw_inode);
499 if (err)
500 goto failed_unmap;
501
502 if (S_ISREG(inode->i_mode)) {
503 inode->i_op = &nilfs_file_inode_operations;
504 inode->i_fop = &nilfs_file_operations;
505 inode->i_mapping->a_ops = &nilfs_aops;
506 } else if (S_ISDIR(inode->i_mode)) {
507 inode->i_op = &nilfs_dir_inode_operations;
508 inode->i_fop = &nilfs_dir_operations;
509 inode->i_mapping->a_ops = &nilfs_aops;
510 } else if (S_ISLNK(inode->i_mode)) {
511 inode->i_op = &nilfs_symlink_inode_operations;
512 inode_nohighmem(inode);
513 inode->i_mapping->a_ops = &nilfs_aops;
514 } else {
515 inode->i_op = &nilfs_special_inode_operations;
516 init_special_inode(
517 inode, inode->i_mode,
518 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
519 }
520 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
521 brelse(bh);
522 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
523 nilfs_set_inode_flags(inode);
524 mapping_set_gfp_mask(inode->i_mapping,
525 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
526 return 0;
527
528 failed_unmap:
529 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
530 brelse(bh);
531
532 bad_inode:
533 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
534 return err;
535}
536
537static int nilfs_iget_test(struct inode *inode, void *opaque)
538{
539 struct nilfs_iget_args *args = opaque;
540 struct nilfs_inode_info *ii;
541
542 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
543 return 0;
544
545 ii = NILFS_I(inode);
546 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
547 return !args->for_gc;
548
549 return args->for_gc && args->cno == ii->i_cno;
550}
551
552static int nilfs_iget_set(struct inode *inode, void *opaque)
553{
554 struct nilfs_iget_args *args = opaque;
555
556 inode->i_ino = args->ino;
557 if (args->for_gc) {
558 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
559 NILFS_I(inode)->i_cno = args->cno;
560 NILFS_I(inode)->i_root = NULL;
561 } else {
562 if (args->root && args->ino == NILFS_ROOT_INO)
563 nilfs_get_root(args->root);
564 NILFS_I(inode)->i_root = args->root;
565 }
566 return 0;
567}
568
569struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
570 unsigned long ino)
571{
572 struct nilfs_iget_args args = {
573 .ino = ino, .root = root, .cno = 0, .for_gc = 0
574 };
575
576 return ilookup5(sb, ino, nilfs_iget_test, &args);
577}
578
579struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
580 unsigned long ino)
581{
582 struct nilfs_iget_args args = {
583 .ino = ino, .root = root, .cno = 0, .for_gc = 0
584 };
585
586 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
587}
588
589struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
590 unsigned long ino)
591{
592 struct inode *inode;
593 int err;
594
595 inode = nilfs_iget_locked(sb, root, ino);
596 if (unlikely(!inode))
597 return ERR_PTR(-ENOMEM);
598 if (!(inode->i_state & I_NEW))
599 return inode;
600
601 err = __nilfs_read_inode(sb, root, ino, inode);
602 if (unlikely(err)) {
603 iget_failed(inode);
604 return ERR_PTR(err);
605 }
606 unlock_new_inode(inode);
607 return inode;
608}
609
610struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
611 __u64 cno)
612{
613 struct nilfs_iget_args args = {
614 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
615 };
616 struct inode *inode;
617 int err;
618
619 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
620 if (unlikely(!inode))
621 return ERR_PTR(-ENOMEM);
622 if (!(inode->i_state & I_NEW))
623 return inode;
624
625 err = nilfs_init_gcinode(inode);
626 if (unlikely(err)) {
627 iget_failed(inode);
628 return ERR_PTR(err);
629 }
630 unlock_new_inode(inode);
631 return inode;
632}
633
634void nilfs_write_inode_common(struct inode *inode,
635 struct nilfs_inode *raw_inode, int has_bmap)
636{
637 struct nilfs_inode_info *ii = NILFS_I(inode);
638
639 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
640 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
641 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
642 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
643 raw_inode->i_size = cpu_to_le64(inode->i_size);
644 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
645 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
646 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
647 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
648 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
649
650 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
651 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
652
653 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
654 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
655
656 /* zero-fill unused portion in the case of super root block */
657 raw_inode->i_xattr = 0;
658 raw_inode->i_pad = 0;
659 memset((void *)raw_inode + sizeof(*raw_inode), 0,
660 nilfs->ns_inode_size - sizeof(*raw_inode));
661 }
662
663 if (has_bmap)
664 nilfs_bmap_write(ii->i_bmap, raw_inode);
665 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
666 raw_inode->i_device_code =
667 cpu_to_le64(huge_encode_dev(inode->i_rdev));
668 /*
669 * When extending inode, nilfs->ns_inode_size should be checked
670 * for substitutions of appended fields.
671 */
672}
673
674void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
675{
676 ino_t ino = inode->i_ino;
677 struct nilfs_inode_info *ii = NILFS_I(inode);
678 struct inode *ifile = ii->i_root->ifile;
679 struct nilfs_inode *raw_inode;
680
681 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
682
683 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
684 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
685 if (flags & I_DIRTY_DATASYNC)
686 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
687
688 nilfs_write_inode_common(inode, raw_inode, 0);
689 /*
690 * XXX: call with has_bmap = 0 is a workaround to avoid
691 * deadlock of bmap. This delays update of i_bmap to just
692 * before writing.
693 */
694
695 nilfs_ifile_unmap_inode(ifile, ino, ibh);
696}
697
698#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
699
700static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
701 unsigned long from)
702{
703 __u64 b;
704 int ret;
705
706 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
707 return;
708repeat:
709 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
710 if (ret == -ENOENT)
711 return;
712 else if (ret < 0)
713 goto failed;
714
715 if (b < from)
716 return;
717
718 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
719 ret = nilfs_bmap_truncate(ii->i_bmap, b);
720 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
721 if (!ret || (ret == -ENOMEM &&
722 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
723 goto repeat;
724
725failed:
726 nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
727 "error %d truncating bmap (ino=%lu)", ret,
728 ii->vfs_inode.i_ino);
729}
730
731void nilfs_truncate(struct inode *inode)
732{
733 unsigned long blkoff;
734 unsigned int blocksize;
735 struct nilfs_transaction_info ti;
736 struct super_block *sb = inode->i_sb;
737 struct nilfs_inode_info *ii = NILFS_I(inode);
738
739 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
740 return;
741 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
742 return;
743
744 blocksize = sb->s_blocksize;
745 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
746 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
747
748 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
749
750 nilfs_truncate_bmap(ii, blkoff);
751
752 inode->i_mtime = inode->i_ctime = current_time(inode);
753 if (IS_SYNC(inode))
754 nilfs_set_transaction_flag(NILFS_TI_SYNC);
755
756 nilfs_mark_inode_dirty(inode);
757 nilfs_set_file_dirty(inode, 0);
758 nilfs_transaction_commit(sb);
759 /*
760 * May construct a logical segment and may fail in sync mode.
761 * But truncate has no return value.
762 */
763}
764
765static void nilfs_clear_inode(struct inode *inode)
766{
767 struct nilfs_inode_info *ii = NILFS_I(inode);
768
769 /*
770 * Free resources allocated in nilfs_read_inode(), here.
771 */
772 BUG_ON(!list_empty(&ii->i_dirty));
773 brelse(ii->i_bh);
774 ii->i_bh = NULL;
775
776 if (nilfs_is_metadata_file_inode(inode))
777 nilfs_mdt_clear(inode);
778
779 if (test_bit(NILFS_I_BMAP, &ii->i_state))
780 nilfs_bmap_clear(ii->i_bmap);
781
782 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
783
784 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
785 nilfs_put_root(ii->i_root);
786}
787
788void nilfs_evict_inode(struct inode *inode)
789{
790 struct nilfs_transaction_info ti;
791 struct super_block *sb = inode->i_sb;
792 struct nilfs_inode_info *ii = NILFS_I(inode);
793 int ret;
794
795 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
796 truncate_inode_pages_final(&inode->i_data);
797 clear_inode(inode);
798 nilfs_clear_inode(inode);
799 return;
800 }
801 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
802
803 truncate_inode_pages_final(&inode->i_data);
804
805 /* TODO: some of the following operations may fail. */
806 nilfs_truncate_bmap(ii, 0);
807 nilfs_mark_inode_dirty(inode);
808 clear_inode(inode);
809
810 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
811 if (!ret)
812 atomic64_dec(&ii->i_root->inodes_count);
813
814 nilfs_clear_inode(inode);
815
816 if (IS_SYNC(inode))
817 nilfs_set_transaction_flag(NILFS_TI_SYNC);
818 nilfs_transaction_commit(sb);
819 /*
820 * May construct a logical segment and may fail in sync mode.
821 * But delete_inode has no return value.
822 */
823}
824
825int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
826{
827 struct nilfs_transaction_info ti;
828 struct inode *inode = d_inode(dentry);
829 struct super_block *sb = inode->i_sb;
830 int err;
831
832 err = setattr_prepare(dentry, iattr);
833 if (err)
834 return err;
835
836 err = nilfs_transaction_begin(sb, &ti, 0);
837 if (unlikely(err))
838 return err;
839
840 if ((iattr->ia_valid & ATTR_SIZE) &&
841 iattr->ia_size != i_size_read(inode)) {
842 inode_dio_wait(inode);
843 truncate_setsize(inode, iattr->ia_size);
844 nilfs_truncate(inode);
845 }
846
847 setattr_copy(inode, iattr);
848 mark_inode_dirty(inode);
849
850 if (iattr->ia_valid & ATTR_MODE) {
851 err = nilfs_acl_chmod(inode);
852 if (unlikely(err))
853 goto out_err;
854 }
855
856 return nilfs_transaction_commit(sb);
857
858out_err:
859 nilfs_transaction_abort(sb);
860 return err;
861}
862
863int nilfs_permission(struct inode *inode, int mask)
864{
865 struct nilfs_root *root = NILFS_I(inode)->i_root;
866
867 if ((mask & MAY_WRITE) && root &&
868 root->cno != NILFS_CPTREE_CURRENT_CNO)
869 return -EROFS; /* snapshot is not writable */
870
871 return generic_permission(inode, mask);
872}
873
874int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
875{
876 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
877 struct nilfs_inode_info *ii = NILFS_I(inode);
878 int err;
879
880 spin_lock(&nilfs->ns_inode_lock);
881 if (ii->i_bh == NULL) {
882 spin_unlock(&nilfs->ns_inode_lock);
883 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
884 inode->i_ino, pbh);
885 if (unlikely(err))
886 return err;
887 spin_lock(&nilfs->ns_inode_lock);
888 if (ii->i_bh == NULL)
889 ii->i_bh = *pbh;
890 else {
891 brelse(*pbh);
892 *pbh = ii->i_bh;
893 }
894 } else
895 *pbh = ii->i_bh;
896
897 get_bh(*pbh);
898 spin_unlock(&nilfs->ns_inode_lock);
899 return 0;
900}
901
902int nilfs_inode_dirty(struct inode *inode)
903{
904 struct nilfs_inode_info *ii = NILFS_I(inode);
905 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
906 int ret = 0;
907
908 if (!list_empty(&ii->i_dirty)) {
909 spin_lock(&nilfs->ns_inode_lock);
910 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
911 test_bit(NILFS_I_BUSY, &ii->i_state);
912 spin_unlock(&nilfs->ns_inode_lock);
913 }
914 return ret;
915}
916
917int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
918{
919 struct nilfs_inode_info *ii = NILFS_I(inode);
920 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
921
922 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
923
924 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
925 return 0;
926
927 spin_lock(&nilfs->ns_inode_lock);
928 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
929 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
930 /*
931 * Because this routine may race with nilfs_dispose_list(),
932 * we have to check NILFS_I_QUEUED here, too.
933 */
934 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
935 /*
936 * This will happen when somebody is freeing
937 * this inode.
938 */
939 nilfs_msg(inode->i_sb, KERN_WARNING,
940 "cannot set file dirty (ino=%lu): the file is being freed",
941 inode->i_ino);
942 spin_unlock(&nilfs->ns_inode_lock);
943 return -EINVAL; /*
944 * NILFS_I_DIRTY may remain for
945 * freeing inode.
946 */
947 }
948 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
949 set_bit(NILFS_I_QUEUED, &ii->i_state);
950 }
951 spin_unlock(&nilfs->ns_inode_lock);
952 return 0;
953}
954
955int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
956{
957 struct buffer_head *ibh;
958 int err;
959
960 err = nilfs_load_inode_block(inode, &ibh);
961 if (unlikely(err)) {
962 nilfs_msg(inode->i_sb, KERN_WARNING,
963 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
964 inode->i_ino, err);
965 return err;
966 }
967 nilfs_update_inode(inode, ibh, flags);
968 mark_buffer_dirty(ibh);
969 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
970 brelse(ibh);
971 return 0;
972}
973
974/**
975 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
976 * @inode: inode of the file to be registered.
977 *
978 * nilfs_dirty_inode() loads a inode block containing the specified
979 * @inode and copies data from a nilfs_inode to a corresponding inode
980 * entry in the inode block. This operation is excluded from the segment
981 * construction. This function can be called both as a single operation
982 * and as a part of indivisible file operations.
983 */
984void nilfs_dirty_inode(struct inode *inode, int flags)
985{
986 struct nilfs_transaction_info ti;
987 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
988
989 if (is_bad_inode(inode)) {
990 nilfs_msg(inode->i_sb, KERN_WARNING,
991 "tried to mark bad_inode dirty. ignored.");
992 dump_stack();
993 return;
994 }
995 if (mdi) {
996 nilfs_mdt_mark_dirty(inode);
997 return;
998 }
999 nilfs_transaction_begin(inode->i_sb, &ti, 0);
1000 __nilfs_mark_inode_dirty(inode, flags);
1001 nilfs_transaction_commit(inode->i_sb); /* never fails */
1002}
1003
1004int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1005 __u64 start, __u64 len)
1006{
1007 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1008 __u64 logical = 0, phys = 0, size = 0;
1009 __u32 flags = 0;
1010 loff_t isize;
1011 sector_t blkoff, end_blkoff;
1012 sector_t delalloc_blkoff;
1013 unsigned long delalloc_blklen;
1014 unsigned int blkbits = inode->i_blkbits;
1015 int ret, n;
1016
1017 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1018 if (ret)
1019 return ret;
1020
1021 inode_lock(inode);
1022
1023 isize = i_size_read(inode);
1024
1025 blkoff = start >> blkbits;
1026 end_blkoff = (start + len - 1) >> blkbits;
1027
1028 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1029 &delalloc_blkoff);
1030
1031 do {
1032 __u64 blkphy;
1033 unsigned int maxblocks;
1034
1035 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1036 if (size) {
1037 /* End of the current extent */
1038 ret = fiemap_fill_next_extent(
1039 fieinfo, logical, phys, size, flags);
1040 if (ret)
1041 break;
1042 }
1043 if (blkoff > end_blkoff)
1044 break;
1045
1046 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1047 logical = blkoff << blkbits;
1048 phys = 0;
1049 size = delalloc_blklen << blkbits;
1050
1051 blkoff = delalloc_blkoff + delalloc_blklen;
1052 delalloc_blklen = nilfs_find_uncommitted_extent(
1053 inode, blkoff, &delalloc_blkoff);
1054 continue;
1055 }
1056
1057 /*
1058 * Limit the number of blocks that we look up so as
1059 * not to get into the next delayed allocation extent.
1060 */
1061 maxblocks = INT_MAX;
1062 if (delalloc_blklen)
1063 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1064 maxblocks);
1065 blkphy = 0;
1066
1067 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1068 n = nilfs_bmap_lookup_contig(
1069 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1070 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1071
1072 if (n < 0) {
1073 int past_eof;
1074
1075 if (unlikely(n != -ENOENT))
1076 break; /* error */
1077
1078 /* HOLE */
1079 blkoff++;
1080 past_eof = ((blkoff << blkbits) >= isize);
1081
1082 if (size) {
1083 /* End of the current extent */
1084
1085 if (past_eof)
1086 flags |= FIEMAP_EXTENT_LAST;
1087
1088 ret = fiemap_fill_next_extent(
1089 fieinfo, logical, phys, size, flags);
1090 if (ret)
1091 break;
1092 size = 0;
1093 }
1094 if (blkoff > end_blkoff || past_eof)
1095 break;
1096 } else {
1097 if (size) {
1098 if (phys && blkphy << blkbits == phys + size) {
1099 /* The current extent goes on */
1100 size += n << blkbits;
1101 } else {
1102 /* Terminate the current extent */
1103 ret = fiemap_fill_next_extent(
1104 fieinfo, logical, phys, size,
1105 flags);
1106 if (ret || blkoff > end_blkoff)
1107 break;
1108
1109 /* Start another extent */
1110 flags = FIEMAP_EXTENT_MERGED;
1111 logical = blkoff << blkbits;
1112 phys = blkphy << blkbits;
1113 size = n << blkbits;
1114 }
1115 } else {
1116 /* Start a new extent */
1117 flags = FIEMAP_EXTENT_MERGED;
1118 logical = blkoff << blkbits;
1119 phys = blkphy << blkbits;
1120 size = n << blkbits;
1121 }
1122 blkoff += n;
1123 }
1124 cond_resched();
1125 } while (true);
1126
1127 /* If ret is 1 then we just hit the end of the extent array */
1128 if (ret == 1)
1129 ret = 0;
1130
1131 inode_unlock(inode);
1132 return ret;
1133}
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * inode.c - NILFS inode operations.
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Ryusuke Konishi.
8 *
9 */
10
11#include <linux/buffer_head.h>
12#include <linux/gfp.h>
13#include <linux/mpage.h>
14#include <linux/pagemap.h>
15#include <linux/writeback.h>
16#include <linux/uio.h>
17#include "nilfs.h"
18#include "btnode.h"
19#include "segment.h"
20#include "page.h"
21#include "mdt.h"
22#include "cpfile.h"
23#include "ifile.h"
24
25/**
26 * struct nilfs_iget_args - arguments used during comparison between inodes
27 * @ino: inode number
28 * @cno: checkpoint number
29 * @root: pointer on NILFS root object (mounted checkpoint)
30 * @for_gc: inode for GC flag
31 */
32struct nilfs_iget_args {
33 u64 ino;
34 __u64 cno;
35 struct nilfs_root *root;
36 int for_gc;
37};
38
39static int nilfs_iget_test(struct inode *inode, void *opaque);
40
41void nilfs_inode_add_blocks(struct inode *inode, int n)
42{
43 struct nilfs_root *root = NILFS_I(inode)->i_root;
44
45 inode_add_bytes(inode, i_blocksize(inode) * n);
46 if (root)
47 atomic64_add(n, &root->blocks_count);
48}
49
50void nilfs_inode_sub_blocks(struct inode *inode, int n)
51{
52 struct nilfs_root *root = NILFS_I(inode)->i_root;
53
54 inode_sub_bytes(inode, i_blocksize(inode) * n);
55 if (root)
56 atomic64_sub(n, &root->blocks_count);
57}
58
59/**
60 * nilfs_get_block() - get a file block on the filesystem (callback function)
61 * @inode - inode struct of the target file
62 * @blkoff - file block number
63 * @bh_result - buffer head to be mapped on
64 * @create - indicate whether allocating the block or not when it has not
65 * been allocated yet.
66 *
67 * This function does not issue actual read request of the specified data
68 * block. It is done by VFS.
69 */
70int nilfs_get_block(struct inode *inode, sector_t blkoff,
71 struct buffer_head *bh_result, int create)
72{
73 struct nilfs_inode_info *ii = NILFS_I(inode);
74 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
75 __u64 blknum = 0;
76 int err = 0, ret;
77 unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
78
79 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
80 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
81 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
82 if (ret >= 0) { /* found */
83 map_bh(bh_result, inode->i_sb, blknum);
84 if (ret > 0)
85 bh_result->b_size = (ret << inode->i_blkbits);
86 goto out;
87 }
88 /* data block was not found */
89 if (ret == -ENOENT && create) {
90 struct nilfs_transaction_info ti;
91
92 bh_result->b_blocknr = 0;
93 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
94 if (unlikely(err))
95 goto out;
96 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
97 (unsigned long)bh_result);
98 if (unlikely(err != 0)) {
99 if (err == -EEXIST) {
100 /*
101 * The get_block() function could be called
102 * from multiple callers for an inode.
103 * However, the page having this block must
104 * be locked in this case.
105 */
106 nilfs_msg(inode->i_sb, KERN_WARNING,
107 "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
108 __func__, inode->i_ino,
109 (unsigned long long)blkoff);
110 err = 0;
111 }
112 nilfs_transaction_abort(inode->i_sb);
113 goto out;
114 }
115 nilfs_mark_inode_dirty_sync(inode);
116 nilfs_transaction_commit(inode->i_sb); /* never fails */
117 /* Error handling should be detailed */
118 set_buffer_new(bh_result);
119 set_buffer_delay(bh_result);
120 map_bh(bh_result, inode->i_sb, 0);
121 /* Disk block number must be changed to proper value */
122
123 } else if (ret == -ENOENT) {
124 /*
125 * not found is not error (e.g. hole); must return without
126 * the mapped state flag.
127 */
128 ;
129 } else {
130 err = ret;
131 }
132
133 out:
134 return err;
135}
136
137/**
138 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
139 * address_space_operations.
140 * @file - file struct of the file to be read
141 * @page - the page to be read
142 */
143static int nilfs_readpage(struct file *file, struct page *page)
144{
145 return mpage_readpage(page, nilfs_get_block);
146}
147
148/**
149 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
150 * address_space_operations.
151 * @file - file struct of the file to be read
152 * @mapping - address_space struct used for reading multiple pages
153 * @pages - the pages to be read
154 * @nr_pages - number of pages to be read
155 */
156static int nilfs_readpages(struct file *file, struct address_space *mapping,
157 struct list_head *pages, unsigned int nr_pages)
158{
159 return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
160}
161
162static int nilfs_writepages(struct address_space *mapping,
163 struct writeback_control *wbc)
164{
165 struct inode *inode = mapping->host;
166 int err = 0;
167
168 if (sb_rdonly(inode->i_sb)) {
169 nilfs_clear_dirty_pages(mapping, false);
170 return -EROFS;
171 }
172
173 if (wbc->sync_mode == WB_SYNC_ALL)
174 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
175 wbc->range_start,
176 wbc->range_end);
177 return err;
178}
179
180static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
181{
182 struct inode *inode = page->mapping->host;
183 int err;
184
185 if (sb_rdonly(inode->i_sb)) {
186 /*
187 * It means that filesystem was remounted in read-only
188 * mode because of error or metadata corruption. But we
189 * have dirty pages that try to be flushed in background.
190 * So, here we simply discard this dirty page.
191 */
192 nilfs_clear_dirty_page(page, false);
193 unlock_page(page);
194 return -EROFS;
195 }
196
197 redirty_page_for_writepage(wbc, page);
198 unlock_page(page);
199
200 if (wbc->sync_mode == WB_SYNC_ALL) {
201 err = nilfs_construct_segment(inode->i_sb);
202 if (unlikely(err))
203 return err;
204 } else if (wbc->for_reclaim)
205 nilfs_flush_segment(inode->i_sb, inode->i_ino);
206
207 return 0;
208}
209
210static int nilfs_set_page_dirty(struct page *page)
211{
212 struct inode *inode = page->mapping->host;
213 int ret = __set_page_dirty_nobuffers(page);
214
215 if (page_has_buffers(page)) {
216 unsigned int nr_dirty = 0;
217 struct buffer_head *bh, *head;
218
219 /*
220 * This page is locked by callers, and no other thread
221 * concurrently marks its buffers dirty since they are
222 * only dirtied through routines in fs/buffer.c in
223 * which call sites of mark_buffer_dirty are protected
224 * by page lock.
225 */
226 bh = head = page_buffers(page);
227 do {
228 /* Do not mark hole blocks dirty */
229 if (buffer_dirty(bh) || !buffer_mapped(bh))
230 continue;
231
232 set_buffer_dirty(bh);
233 nr_dirty++;
234 } while (bh = bh->b_this_page, bh != head);
235
236 if (nr_dirty)
237 nilfs_set_file_dirty(inode, nr_dirty);
238 } else if (ret) {
239 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
240
241 nilfs_set_file_dirty(inode, nr_dirty);
242 }
243 return ret;
244}
245
246void nilfs_write_failed(struct address_space *mapping, loff_t to)
247{
248 struct inode *inode = mapping->host;
249
250 if (to > inode->i_size) {
251 truncate_pagecache(inode, inode->i_size);
252 nilfs_truncate(inode);
253 }
254}
255
256static int nilfs_write_begin(struct file *file, struct address_space *mapping,
257 loff_t pos, unsigned len, unsigned flags,
258 struct page **pagep, void **fsdata)
259
260{
261 struct inode *inode = mapping->host;
262 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
263
264 if (unlikely(err))
265 return err;
266
267 err = block_write_begin(mapping, pos, len, flags, pagep,
268 nilfs_get_block);
269 if (unlikely(err)) {
270 nilfs_write_failed(mapping, pos + len);
271 nilfs_transaction_abort(inode->i_sb);
272 }
273 return err;
274}
275
276static int nilfs_write_end(struct file *file, struct address_space *mapping,
277 loff_t pos, unsigned len, unsigned copied,
278 struct page *page, void *fsdata)
279{
280 struct inode *inode = mapping->host;
281 unsigned int start = pos & (PAGE_SIZE - 1);
282 unsigned int nr_dirty;
283 int err;
284
285 nr_dirty = nilfs_page_count_clean_buffers(page, start,
286 start + copied);
287 copied = generic_write_end(file, mapping, pos, len, copied, page,
288 fsdata);
289 nilfs_set_file_dirty(inode, nr_dirty);
290 err = nilfs_transaction_commit(inode->i_sb);
291 return err ? : copied;
292}
293
294static ssize_t
295nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
296{
297 struct inode *inode = file_inode(iocb->ki_filp);
298
299 if (iov_iter_rw(iter) == WRITE)
300 return 0;
301
302 /* Needs synchronization with the cleaner */
303 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
304}
305
306const struct address_space_operations nilfs_aops = {
307 .writepage = nilfs_writepage,
308 .readpage = nilfs_readpage,
309 .writepages = nilfs_writepages,
310 .set_page_dirty = nilfs_set_page_dirty,
311 .readpages = nilfs_readpages,
312 .write_begin = nilfs_write_begin,
313 .write_end = nilfs_write_end,
314 /* .releasepage = nilfs_releasepage, */
315 .invalidatepage = block_invalidatepage,
316 .direct_IO = nilfs_direct_IO,
317 .is_partially_uptodate = block_is_partially_uptodate,
318};
319
320static int nilfs_insert_inode_locked(struct inode *inode,
321 struct nilfs_root *root,
322 unsigned long ino)
323{
324 struct nilfs_iget_args args = {
325 .ino = ino, .root = root, .cno = 0, .for_gc = 0
326 };
327
328 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
329}
330
331struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
332{
333 struct super_block *sb = dir->i_sb;
334 struct the_nilfs *nilfs = sb->s_fs_info;
335 struct inode *inode;
336 struct nilfs_inode_info *ii;
337 struct nilfs_root *root;
338 int err = -ENOMEM;
339 ino_t ino;
340
341 inode = new_inode(sb);
342 if (unlikely(!inode))
343 goto failed;
344
345 mapping_set_gfp_mask(inode->i_mapping,
346 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
347
348 root = NILFS_I(dir)->i_root;
349 ii = NILFS_I(inode);
350 ii->i_state = BIT(NILFS_I_NEW);
351 ii->i_root = root;
352
353 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
354 if (unlikely(err))
355 goto failed_ifile_create_inode;
356 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
357
358 atomic64_inc(&root->inodes_count);
359 inode_init_owner(inode, dir, mode);
360 inode->i_ino = ino;
361 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
362
363 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
364 err = nilfs_bmap_read(ii->i_bmap, NULL);
365 if (err < 0)
366 goto failed_after_creation;
367
368 set_bit(NILFS_I_BMAP, &ii->i_state);
369 /* No lock is needed; iget() ensures it. */
370 }
371
372 ii->i_flags = nilfs_mask_flags(
373 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
374
375 /* ii->i_file_acl = 0; */
376 /* ii->i_dir_acl = 0; */
377 ii->i_dir_start_lookup = 0;
378 nilfs_set_inode_flags(inode);
379 spin_lock(&nilfs->ns_next_gen_lock);
380 inode->i_generation = nilfs->ns_next_generation++;
381 spin_unlock(&nilfs->ns_next_gen_lock);
382 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
383 err = -EIO;
384 goto failed_after_creation;
385 }
386
387 err = nilfs_init_acl(inode, dir);
388 if (unlikely(err))
389 /*
390 * Never occur. When supporting nilfs_init_acl(),
391 * proper cancellation of above jobs should be considered.
392 */
393 goto failed_after_creation;
394
395 return inode;
396
397 failed_after_creation:
398 clear_nlink(inode);
399 unlock_new_inode(inode);
400 iput(inode); /*
401 * raw_inode will be deleted through
402 * nilfs_evict_inode().
403 */
404 goto failed;
405
406 failed_ifile_create_inode:
407 make_bad_inode(inode);
408 iput(inode);
409 failed:
410 return ERR_PTR(err);
411}
412
413void nilfs_set_inode_flags(struct inode *inode)
414{
415 unsigned int flags = NILFS_I(inode)->i_flags;
416 unsigned int new_fl = 0;
417
418 if (flags & FS_SYNC_FL)
419 new_fl |= S_SYNC;
420 if (flags & FS_APPEND_FL)
421 new_fl |= S_APPEND;
422 if (flags & FS_IMMUTABLE_FL)
423 new_fl |= S_IMMUTABLE;
424 if (flags & FS_NOATIME_FL)
425 new_fl |= S_NOATIME;
426 if (flags & FS_DIRSYNC_FL)
427 new_fl |= S_DIRSYNC;
428 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
429 S_NOATIME | S_DIRSYNC);
430}
431
432int nilfs_read_inode_common(struct inode *inode,
433 struct nilfs_inode *raw_inode)
434{
435 struct nilfs_inode_info *ii = NILFS_I(inode);
436 int err;
437
438 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
439 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
440 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
441 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
442 inode->i_size = le64_to_cpu(raw_inode->i_size);
443 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
444 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
445 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
446 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
447 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
448 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
449 if (inode->i_nlink == 0)
450 return -ESTALE; /* this inode is deleted */
451
452 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
453 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
454#if 0
455 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
456 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
457 0 : le32_to_cpu(raw_inode->i_dir_acl);
458#endif
459 ii->i_dir_start_lookup = 0;
460 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
461
462 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
463 S_ISLNK(inode->i_mode)) {
464 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
465 if (err < 0)
466 return err;
467 set_bit(NILFS_I_BMAP, &ii->i_state);
468 /* No lock is needed; iget() ensures it. */
469 }
470 return 0;
471}
472
473static int __nilfs_read_inode(struct super_block *sb,
474 struct nilfs_root *root, unsigned long ino,
475 struct inode *inode)
476{
477 struct the_nilfs *nilfs = sb->s_fs_info;
478 struct buffer_head *bh;
479 struct nilfs_inode *raw_inode;
480 int err;
481
482 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
483 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
484 if (unlikely(err))
485 goto bad_inode;
486
487 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
488
489 err = nilfs_read_inode_common(inode, raw_inode);
490 if (err)
491 goto failed_unmap;
492
493 if (S_ISREG(inode->i_mode)) {
494 inode->i_op = &nilfs_file_inode_operations;
495 inode->i_fop = &nilfs_file_operations;
496 inode->i_mapping->a_ops = &nilfs_aops;
497 } else if (S_ISDIR(inode->i_mode)) {
498 inode->i_op = &nilfs_dir_inode_operations;
499 inode->i_fop = &nilfs_dir_operations;
500 inode->i_mapping->a_ops = &nilfs_aops;
501 } else if (S_ISLNK(inode->i_mode)) {
502 inode->i_op = &nilfs_symlink_inode_operations;
503 inode_nohighmem(inode);
504 inode->i_mapping->a_ops = &nilfs_aops;
505 } else {
506 inode->i_op = &nilfs_special_inode_operations;
507 init_special_inode(
508 inode, inode->i_mode,
509 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
510 }
511 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
512 brelse(bh);
513 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
514 nilfs_set_inode_flags(inode);
515 mapping_set_gfp_mask(inode->i_mapping,
516 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
517 return 0;
518
519 failed_unmap:
520 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
521 brelse(bh);
522
523 bad_inode:
524 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
525 return err;
526}
527
528static int nilfs_iget_test(struct inode *inode, void *opaque)
529{
530 struct nilfs_iget_args *args = opaque;
531 struct nilfs_inode_info *ii;
532
533 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
534 return 0;
535
536 ii = NILFS_I(inode);
537 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
538 return !args->for_gc;
539
540 return args->for_gc && args->cno == ii->i_cno;
541}
542
543static int nilfs_iget_set(struct inode *inode, void *opaque)
544{
545 struct nilfs_iget_args *args = opaque;
546
547 inode->i_ino = args->ino;
548 if (args->for_gc) {
549 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
550 NILFS_I(inode)->i_cno = args->cno;
551 NILFS_I(inode)->i_root = NULL;
552 } else {
553 if (args->root && args->ino == NILFS_ROOT_INO)
554 nilfs_get_root(args->root);
555 NILFS_I(inode)->i_root = args->root;
556 }
557 return 0;
558}
559
560struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
561 unsigned long ino)
562{
563 struct nilfs_iget_args args = {
564 .ino = ino, .root = root, .cno = 0, .for_gc = 0
565 };
566
567 return ilookup5(sb, ino, nilfs_iget_test, &args);
568}
569
570struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
571 unsigned long ino)
572{
573 struct nilfs_iget_args args = {
574 .ino = ino, .root = root, .cno = 0, .for_gc = 0
575 };
576
577 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
578}
579
580struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
581 unsigned long ino)
582{
583 struct inode *inode;
584 int err;
585
586 inode = nilfs_iget_locked(sb, root, ino);
587 if (unlikely(!inode))
588 return ERR_PTR(-ENOMEM);
589 if (!(inode->i_state & I_NEW))
590 return inode;
591
592 err = __nilfs_read_inode(sb, root, ino, inode);
593 if (unlikely(err)) {
594 iget_failed(inode);
595 return ERR_PTR(err);
596 }
597 unlock_new_inode(inode);
598 return inode;
599}
600
601struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
602 __u64 cno)
603{
604 struct nilfs_iget_args args = {
605 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
606 };
607 struct inode *inode;
608 int err;
609
610 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
611 if (unlikely(!inode))
612 return ERR_PTR(-ENOMEM);
613 if (!(inode->i_state & I_NEW))
614 return inode;
615
616 err = nilfs_init_gcinode(inode);
617 if (unlikely(err)) {
618 iget_failed(inode);
619 return ERR_PTR(err);
620 }
621 unlock_new_inode(inode);
622 return inode;
623}
624
625void nilfs_write_inode_common(struct inode *inode,
626 struct nilfs_inode *raw_inode, int has_bmap)
627{
628 struct nilfs_inode_info *ii = NILFS_I(inode);
629
630 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
631 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
632 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
633 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
634 raw_inode->i_size = cpu_to_le64(inode->i_size);
635 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
636 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
637 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
638 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
639 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
640
641 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
642 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
643
644 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
645 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
646
647 /* zero-fill unused portion in the case of super root block */
648 raw_inode->i_xattr = 0;
649 raw_inode->i_pad = 0;
650 memset((void *)raw_inode + sizeof(*raw_inode), 0,
651 nilfs->ns_inode_size - sizeof(*raw_inode));
652 }
653
654 if (has_bmap)
655 nilfs_bmap_write(ii->i_bmap, raw_inode);
656 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
657 raw_inode->i_device_code =
658 cpu_to_le64(huge_encode_dev(inode->i_rdev));
659 /*
660 * When extending inode, nilfs->ns_inode_size should be checked
661 * for substitutions of appended fields.
662 */
663}
664
665void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
666{
667 ino_t ino = inode->i_ino;
668 struct nilfs_inode_info *ii = NILFS_I(inode);
669 struct inode *ifile = ii->i_root->ifile;
670 struct nilfs_inode *raw_inode;
671
672 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
673
674 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
675 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
676 if (flags & I_DIRTY_DATASYNC)
677 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
678
679 nilfs_write_inode_common(inode, raw_inode, 0);
680 /*
681 * XXX: call with has_bmap = 0 is a workaround to avoid
682 * deadlock of bmap. This delays update of i_bmap to just
683 * before writing.
684 */
685
686 nilfs_ifile_unmap_inode(ifile, ino, ibh);
687}
688
689#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
690
691static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
692 unsigned long from)
693{
694 __u64 b;
695 int ret;
696
697 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
698 return;
699repeat:
700 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
701 if (ret == -ENOENT)
702 return;
703 else if (ret < 0)
704 goto failed;
705
706 if (b < from)
707 return;
708
709 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
710 ret = nilfs_bmap_truncate(ii->i_bmap, b);
711 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
712 if (!ret || (ret == -ENOMEM &&
713 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
714 goto repeat;
715
716failed:
717 nilfs_msg(ii->vfs_inode.i_sb, KERN_WARNING,
718 "error %d truncating bmap (ino=%lu)", ret,
719 ii->vfs_inode.i_ino);
720}
721
722void nilfs_truncate(struct inode *inode)
723{
724 unsigned long blkoff;
725 unsigned int blocksize;
726 struct nilfs_transaction_info ti;
727 struct super_block *sb = inode->i_sb;
728 struct nilfs_inode_info *ii = NILFS_I(inode);
729
730 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
731 return;
732 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
733 return;
734
735 blocksize = sb->s_blocksize;
736 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
737 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
738
739 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
740
741 nilfs_truncate_bmap(ii, blkoff);
742
743 inode->i_mtime = inode->i_ctime = current_time(inode);
744 if (IS_SYNC(inode))
745 nilfs_set_transaction_flag(NILFS_TI_SYNC);
746
747 nilfs_mark_inode_dirty(inode);
748 nilfs_set_file_dirty(inode, 0);
749 nilfs_transaction_commit(sb);
750 /*
751 * May construct a logical segment and may fail in sync mode.
752 * But truncate has no return value.
753 */
754}
755
756static void nilfs_clear_inode(struct inode *inode)
757{
758 struct nilfs_inode_info *ii = NILFS_I(inode);
759
760 /*
761 * Free resources allocated in nilfs_read_inode(), here.
762 */
763 BUG_ON(!list_empty(&ii->i_dirty));
764 brelse(ii->i_bh);
765 ii->i_bh = NULL;
766
767 if (nilfs_is_metadata_file_inode(inode))
768 nilfs_mdt_clear(inode);
769
770 if (test_bit(NILFS_I_BMAP, &ii->i_state))
771 nilfs_bmap_clear(ii->i_bmap);
772
773 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
774
775 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
776 nilfs_put_root(ii->i_root);
777}
778
779void nilfs_evict_inode(struct inode *inode)
780{
781 struct nilfs_transaction_info ti;
782 struct super_block *sb = inode->i_sb;
783 struct nilfs_inode_info *ii = NILFS_I(inode);
784 int ret;
785
786 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
787 truncate_inode_pages_final(&inode->i_data);
788 clear_inode(inode);
789 nilfs_clear_inode(inode);
790 return;
791 }
792 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
793
794 truncate_inode_pages_final(&inode->i_data);
795
796 /* TODO: some of the following operations may fail. */
797 nilfs_truncate_bmap(ii, 0);
798 nilfs_mark_inode_dirty(inode);
799 clear_inode(inode);
800
801 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
802 if (!ret)
803 atomic64_dec(&ii->i_root->inodes_count);
804
805 nilfs_clear_inode(inode);
806
807 if (IS_SYNC(inode))
808 nilfs_set_transaction_flag(NILFS_TI_SYNC);
809 nilfs_transaction_commit(sb);
810 /*
811 * May construct a logical segment and may fail in sync mode.
812 * But delete_inode has no return value.
813 */
814}
815
816int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
817{
818 struct nilfs_transaction_info ti;
819 struct inode *inode = d_inode(dentry);
820 struct super_block *sb = inode->i_sb;
821 int err;
822
823 err = setattr_prepare(dentry, iattr);
824 if (err)
825 return err;
826
827 err = nilfs_transaction_begin(sb, &ti, 0);
828 if (unlikely(err))
829 return err;
830
831 if ((iattr->ia_valid & ATTR_SIZE) &&
832 iattr->ia_size != i_size_read(inode)) {
833 inode_dio_wait(inode);
834 truncate_setsize(inode, iattr->ia_size);
835 nilfs_truncate(inode);
836 }
837
838 setattr_copy(inode, iattr);
839 mark_inode_dirty(inode);
840
841 if (iattr->ia_valid & ATTR_MODE) {
842 err = nilfs_acl_chmod(inode);
843 if (unlikely(err))
844 goto out_err;
845 }
846
847 return nilfs_transaction_commit(sb);
848
849out_err:
850 nilfs_transaction_abort(sb);
851 return err;
852}
853
854int nilfs_permission(struct inode *inode, int mask)
855{
856 struct nilfs_root *root = NILFS_I(inode)->i_root;
857
858 if ((mask & MAY_WRITE) && root &&
859 root->cno != NILFS_CPTREE_CURRENT_CNO)
860 return -EROFS; /* snapshot is not writable */
861
862 return generic_permission(inode, mask);
863}
864
865int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
866{
867 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
868 struct nilfs_inode_info *ii = NILFS_I(inode);
869 int err;
870
871 spin_lock(&nilfs->ns_inode_lock);
872 if (ii->i_bh == NULL) {
873 spin_unlock(&nilfs->ns_inode_lock);
874 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
875 inode->i_ino, pbh);
876 if (unlikely(err))
877 return err;
878 spin_lock(&nilfs->ns_inode_lock);
879 if (ii->i_bh == NULL)
880 ii->i_bh = *pbh;
881 else {
882 brelse(*pbh);
883 *pbh = ii->i_bh;
884 }
885 } else
886 *pbh = ii->i_bh;
887
888 get_bh(*pbh);
889 spin_unlock(&nilfs->ns_inode_lock);
890 return 0;
891}
892
893int nilfs_inode_dirty(struct inode *inode)
894{
895 struct nilfs_inode_info *ii = NILFS_I(inode);
896 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
897 int ret = 0;
898
899 if (!list_empty(&ii->i_dirty)) {
900 spin_lock(&nilfs->ns_inode_lock);
901 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
902 test_bit(NILFS_I_BUSY, &ii->i_state);
903 spin_unlock(&nilfs->ns_inode_lock);
904 }
905 return ret;
906}
907
908int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
909{
910 struct nilfs_inode_info *ii = NILFS_I(inode);
911 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
912
913 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
914
915 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
916 return 0;
917
918 spin_lock(&nilfs->ns_inode_lock);
919 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
920 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
921 /*
922 * Because this routine may race with nilfs_dispose_list(),
923 * we have to check NILFS_I_QUEUED here, too.
924 */
925 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
926 /*
927 * This will happen when somebody is freeing
928 * this inode.
929 */
930 nilfs_msg(inode->i_sb, KERN_WARNING,
931 "cannot set file dirty (ino=%lu): the file is being freed",
932 inode->i_ino);
933 spin_unlock(&nilfs->ns_inode_lock);
934 return -EINVAL; /*
935 * NILFS_I_DIRTY may remain for
936 * freeing inode.
937 */
938 }
939 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
940 set_bit(NILFS_I_QUEUED, &ii->i_state);
941 }
942 spin_unlock(&nilfs->ns_inode_lock);
943 return 0;
944}
945
946int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
947{
948 struct buffer_head *ibh;
949 int err;
950
951 err = nilfs_load_inode_block(inode, &ibh);
952 if (unlikely(err)) {
953 nilfs_msg(inode->i_sb, KERN_WARNING,
954 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
955 inode->i_ino, err);
956 return err;
957 }
958 nilfs_update_inode(inode, ibh, flags);
959 mark_buffer_dirty(ibh);
960 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
961 brelse(ibh);
962 return 0;
963}
964
965/**
966 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
967 * @inode: inode of the file to be registered.
968 *
969 * nilfs_dirty_inode() loads a inode block containing the specified
970 * @inode and copies data from a nilfs_inode to a corresponding inode
971 * entry in the inode block. This operation is excluded from the segment
972 * construction. This function can be called both as a single operation
973 * and as a part of indivisible file operations.
974 */
975void nilfs_dirty_inode(struct inode *inode, int flags)
976{
977 struct nilfs_transaction_info ti;
978 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
979
980 if (is_bad_inode(inode)) {
981 nilfs_msg(inode->i_sb, KERN_WARNING,
982 "tried to mark bad_inode dirty. ignored.");
983 dump_stack();
984 return;
985 }
986 if (mdi) {
987 nilfs_mdt_mark_dirty(inode);
988 return;
989 }
990 nilfs_transaction_begin(inode->i_sb, &ti, 0);
991 __nilfs_mark_inode_dirty(inode, flags);
992 nilfs_transaction_commit(inode->i_sb); /* never fails */
993}
994
995int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
996 __u64 start, __u64 len)
997{
998 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
999 __u64 logical = 0, phys = 0, size = 0;
1000 __u32 flags = 0;
1001 loff_t isize;
1002 sector_t blkoff, end_blkoff;
1003 sector_t delalloc_blkoff;
1004 unsigned long delalloc_blklen;
1005 unsigned int blkbits = inode->i_blkbits;
1006 int ret, n;
1007
1008 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1009 if (ret)
1010 return ret;
1011
1012 inode_lock(inode);
1013
1014 isize = i_size_read(inode);
1015
1016 blkoff = start >> blkbits;
1017 end_blkoff = (start + len - 1) >> blkbits;
1018
1019 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1020 &delalloc_blkoff);
1021
1022 do {
1023 __u64 blkphy;
1024 unsigned int maxblocks;
1025
1026 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1027 if (size) {
1028 /* End of the current extent */
1029 ret = fiemap_fill_next_extent(
1030 fieinfo, logical, phys, size, flags);
1031 if (ret)
1032 break;
1033 }
1034 if (blkoff > end_blkoff)
1035 break;
1036
1037 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1038 logical = blkoff << blkbits;
1039 phys = 0;
1040 size = delalloc_blklen << blkbits;
1041
1042 blkoff = delalloc_blkoff + delalloc_blklen;
1043 delalloc_blklen = nilfs_find_uncommitted_extent(
1044 inode, blkoff, &delalloc_blkoff);
1045 continue;
1046 }
1047
1048 /*
1049 * Limit the number of blocks that we look up so as
1050 * not to get into the next delayed allocation extent.
1051 */
1052 maxblocks = INT_MAX;
1053 if (delalloc_blklen)
1054 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1055 maxblocks);
1056 blkphy = 0;
1057
1058 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1059 n = nilfs_bmap_lookup_contig(
1060 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1061 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1062
1063 if (n < 0) {
1064 int past_eof;
1065
1066 if (unlikely(n != -ENOENT))
1067 break; /* error */
1068
1069 /* HOLE */
1070 blkoff++;
1071 past_eof = ((blkoff << blkbits) >= isize);
1072
1073 if (size) {
1074 /* End of the current extent */
1075
1076 if (past_eof)
1077 flags |= FIEMAP_EXTENT_LAST;
1078
1079 ret = fiemap_fill_next_extent(
1080 fieinfo, logical, phys, size, flags);
1081 if (ret)
1082 break;
1083 size = 0;
1084 }
1085 if (blkoff > end_blkoff || past_eof)
1086 break;
1087 } else {
1088 if (size) {
1089 if (phys && blkphy << blkbits == phys + size) {
1090 /* The current extent goes on */
1091 size += n << blkbits;
1092 } else {
1093 /* Terminate the current extent */
1094 ret = fiemap_fill_next_extent(
1095 fieinfo, logical, phys, size,
1096 flags);
1097 if (ret || blkoff > end_blkoff)
1098 break;
1099
1100 /* Start another extent */
1101 flags = FIEMAP_EXTENT_MERGED;
1102 logical = blkoff << blkbits;
1103 phys = blkphy << blkbits;
1104 size = n << blkbits;
1105 }
1106 } else {
1107 /* Start a new extent */
1108 flags = FIEMAP_EXTENT_MERGED;
1109 logical = blkoff << blkbits;
1110 phys = blkphy << blkbits;
1111 size = n << blkbits;
1112 }
1113 blkoff += n;
1114 }
1115 cond_resched();
1116 } while (true);
1117
1118 /* If ret is 1 then we just hit the end of the extent array */
1119 if (ret == 1)
1120 ret = 0;
1121
1122 inode_unlock(inode);
1123 return ret;
1124}