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 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
21 *
22 */
23
24#include <linux/buffer_head.h>
25#include <linux/gfp.h>
26#include <linux/mpage.h>
27#include <linux/pagemap.h>
28#include <linux/writeback.h>
29#include <linux/uio.h>
30#include "nilfs.h"
31#include "btnode.h"
32#include "segment.h"
33#include "page.h"
34#include "mdt.h"
35#include "cpfile.h"
36#include "ifile.h"
37
38/**
39 * struct nilfs_iget_args - arguments used during comparison between inodes
40 * @ino: inode number
41 * @cno: checkpoint number
42 * @root: pointer on NILFS root object (mounted checkpoint)
43 * @for_gc: inode for GC flag
44 */
45struct nilfs_iget_args {
46 u64 ino;
47 __u64 cno;
48 struct nilfs_root *root;
49 int for_gc;
50};
51
52static int nilfs_iget_test(struct inode *inode, void *opaque);
53
54void nilfs_inode_add_blocks(struct inode *inode, int n)
55{
56 struct nilfs_root *root = NILFS_I(inode)->i_root;
57
58 inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
59 if (root)
60 atomic64_add(n, &root->blocks_count);
61}
62
63void nilfs_inode_sub_blocks(struct inode *inode, int n)
64{
65 struct nilfs_root *root = NILFS_I(inode)->i_root;
66
67 inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
68 if (root)
69 atomic64_sub(n, &root->blocks_count);
70}
71
72/**
73 * nilfs_get_block() - get a file block on the filesystem (callback function)
74 * @inode - inode struct of the target file
75 * @blkoff - file block number
76 * @bh_result - buffer head to be mapped on
77 * @create - indicate whether allocating the block or not when it has not
78 * been allocated yet.
79 *
80 * This function does not issue actual read request of the specified data
81 * block. It is done by VFS.
82 */
83int nilfs_get_block(struct inode *inode, sector_t blkoff,
84 struct buffer_head *bh_result, int create)
85{
86 struct nilfs_inode_info *ii = NILFS_I(inode);
87 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
88 __u64 blknum = 0;
89 int err = 0, ret;
90 unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
91
92 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
93 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
94 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
95 if (ret >= 0) { /* found */
96 map_bh(bh_result, inode->i_sb, blknum);
97 if (ret > 0)
98 bh_result->b_size = (ret << inode->i_blkbits);
99 goto out;
100 }
101 /* data block was not found */
102 if (ret == -ENOENT && create) {
103 struct nilfs_transaction_info ti;
104
105 bh_result->b_blocknr = 0;
106 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
107 if (unlikely(err))
108 goto out;
109 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
110 (unsigned long)bh_result);
111 if (unlikely(err != 0)) {
112 if (err == -EEXIST) {
113 /*
114 * The get_block() function could be called
115 * from multiple callers for an inode.
116 * However, the page having this block must
117 * be locked in this case.
118 */
119 printk(KERN_WARNING
120 "nilfs_get_block: a race condition "
121 "while inserting a data block. "
122 "(inode number=%lu, file block "
123 "offset=%llu)\n",
124 inode->i_ino,
125 (unsigned long long)blkoff);
126 err = 0;
127 }
128 nilfs_transaction_abort(inode->i_sb);
129 goto out;
130 }
131 nilfs_mark_inode_dirty_sync(inode);
132 nilfs_transaction_commit(inode->i_sb); /* never fails */
133 /* Error handling should be detailed */
134 set_buffer_new(bh_result);
135 set_buffer_delay(bh_result);
136 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
137 to proper value */
138 } else if (ret == -ENOENT) {
139 /* not found is not error (e.g. hole); must return without
140 the mapped state flag. */
141 ;
142 } else {
143 err = ret;
144 }
145
146 out:
147 return err;
148}
149
150/**
151 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
152 * address_space_operations.
153 * @file - file struct of the file to be read
154 * @page - the page to be read
155 */
156static int nilfs_readpage(struct file *file, struct page *page)
157{
158 return mpage_readpage(page, nilfs_get_block);
159}
160
161/**
162 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
163 * address_space_operations.
164 * @file - file struct of the file to be read
165 * @mapping - address_space struct used for reading multiple pages
166 * @pages - the pages to be read
167 * @nr_pages - number of pages to be read
168 */
169static int nilfs_readpages(struct file *file, struct address_space *mapping,
170 struct list_head *pages, unsigned nr_pages)
171{
172 return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
173}
174
175static int nilfs_writepages(struct address_space *mapping,
176 struct writeback_control *wbc)
177{
178 struct inode *inode = mapping->host;
179 int err = 0;
180
181 if (inode->i_sb->s_flags & MS_RDONLY) {
182 nilfs_clear_dirty_pages(mapping, false);
183 return -EROFS;
184 }
185
186 if (wbc->sync_mode == WB_SYNC_ALL)
187 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
188 wbc->range_start,
189 wbc->range_end);
190 return err;
191}
192
193static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
194{
195 struct inode *inode = page->mapping->host;
196 int err;
197
198 if (inode->i_sb->s_flags & MS_RDONLY) {
199 /*
200 * It means that filesystem was remounted in read-only
201 * mode because of error or metadata corruption. But we
202 * have dirty pages that try to be flushed in background.
203 * So, here we simply discard this dirty page.
204 */
205 nilfs_clear_dirty_page(page, false);
206 unlock_page(page);
207 return -EROFS;
208 }
209
210 redirty_page_for_writepage(wbc, page);
211 unlock_page(page);
212
213 if (wbc->sync_mode == WB_SYNC_ALL) {
214 err = nilfs_construct_segment(inode->i_sb);
215 if (unlikely(err))
216 return err;
217 } else if (wbc->for_reclaim)
218 nilfs_flush_segment(inode->i_sb, inode->i_ino);
219
220 return 0;
221}
222
223static int nilfs_set_page_dirty(struct page *page)
224{
225 struct inode *inode = page->mapping->host;
226 int ret = __set_page_dirty_nobuffers(page);
227
228 if (page_has_buffers(page)) {
229 unsigned nr_dirty = 0;
230 struct buffer_head *bh, *head;
231
232 /*
233 * This page is locked by callers, and no other thread
234 * concurrently marks its buffers dirty since they are
235 * only dirtied through routines in fs/buffer.c in
236 * which call sites of mark_buffer_dirty are protected
237 * by page lock.
238 */
239 bh = head = page_buffers(page);
240 do {
241 /* Do not mark hole blocks dirty */
242 if (buffer_dirty(bh) || !buffer_mapped(bh))
243 continue;
244
245 set_buffer_dirty(bh);
246 nr_dirty++;
247 } while (bh = bh->b_this_page, bh != head);
248
249 if (nr_dirty)
250 nilfs_set_file_dirty(inode, nr_dirty);
251 } else if (ret) {
252 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
253
254 nilfs_set_file_dirty(inode, nr_dirty);
255 }
256 return ret;
257}
258
259void nilfs_write_failed(struct address_space *mapping, loff_t to)
260{
261 struct inode *inode = mapping->host;
262
263 if (to > inode->i_size) {
264 truncate_pagecache(inode, inode->i_size);
265 nilfs_truncate(inode);
266 }
267}
268
269static int nilfs_write_begin(struct file *file, struct address_space *mapping,
270 loff_t pos, unsigned len, unsigned flags,
271 struct page **pagep, void **fsdata)
272
273{
274 struct inode *inode = mapping->host;
275 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
276
277 if (unlikely(err))
278 return err;
279
280 err = block_write_begin(mapping, pos, len, flags, pagep,
281 nilfs_get_block);
282 if (unlikely(err)) {
283 nilfs_write_failed(mapping, pos + len);
284 nilfs_transaction_abort(inode->i_sb);
285 }
286 return err;
287}
288
289static int nilfs_write_end(struct file *file, struct address_space *mapping,
290 loff_t pos, unsigned len, unsigned copied,
291 struct page *page, void *fsdata)
292{
293 struct inode *inode = mapping->host;
294 unsigned start = pos & (PAGE_SIZE - 1);
295 unsigned nr_dirty;
296 int err;
297
298 nr_dirty = nilfs_page_count_clean_buffers(page, start,
299 start + copied);
300 copied = generic_write_end(file, mapping, pos, len, copied, page,
301 fsdata);
302 nilfs_set_file_dirty(inode, nr_dirty);
303 err = nilfs_transaction_commit(inode->i_sb);
304 return err ? : copied;
305}
306
307static ssize_t
308nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
309{
310 struct inode *inode = file_inode(iocb->ki_filp);
311
312 if (iov_iter_rw(iter) == WRITE)
313 return 0;
314
315 /* Needs synchronization with the cleaner */
316 return blockdev_direct_IO(iocb, inode, iter, offset, nilfs_get_block);
317}
318
319const struct address_space_operations nilfs_aops = {
320 .writepage = nilfs_writepage,
321 .readpage = nilfs_readpage,
322 .writepages = nilfs_writepages,
323 .set_page_dirty = nilfs_set_page_dirty,
324 .readpages = nilfs_readpages,
325 .write_begin = nilfs_write_begin,
326 .write_end = nilfs_write_end,
327 /* .releasepage = nilfs_releasepage, */
328 .invalidatepage = block_invalidatepage,
329 .direct_IO = nilfs_direct_IO,
330 .is_partially_uptodate = block_is_partially_uptodate,
331};
332
333static int nilfs_insert_inode_locked(struct inode *inode,
334 struct nilfs_root *root,
335 unsigned long ino)
336{
337 struct nilfs_iget_args args = {
338 .ino = ino, .root = root, .cno = 0, .for_gc = 0
339 };
340
341 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
342}
343
344struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
345{
346 struct super_block *sb = dir->i_sb;
347 struct the_nilfs *nilfs = sb->s_fs_info;
348 struct inode *inode;
349 struct nilfs_inode_info *ii;
350 struct nilfs_root *root;
351 int err = -ENOMEM;
352 ino_t ino;
353
354 inode = new_inode(sb);
355 if (unlikely(!inode))
356 goto failed;
357
358 mapping_set_gfp_mask(inode->i_mapping,
359 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
360
361 root = NILFS_I(dir)->i_root;
362 ii = NILFS_I(inode);
363 ii->i_state = 1 << NILFS_I_NEW;
364 ii->i_root = root;
365
366 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
367 if (unlikely(err))
368 goto failed_ifile_create_inode;
369 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
370
371 atomic64_inc(&root->inodes_count);
372 inode_init_owner(inode, dir, mode);
373 inode->i_ino = ino;
374 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
375
376 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
377 err = nilfs_bmap_read(ii->i_bmap, NULL);
378 if (err < 0)
379 goto failed_after_creation;
380
381 set_bit(NILFS_I_BMAP, &ii->i_state);
382 /* No lock is needed; iget() ensures it. */
383 }
384
385 ii->i_flags = nilfs_mask_flags(
386 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
387
388 /* ii->i_file_acl = 0; */
389 /* ii->i_dir_acl = 0; */
390 ii->i_dir_start_lookup = 0;
391 nilfs_set_inode_flags(inode);
392 spin_lock(&nilfs->ns_next_gen_lock);
393 inode->i_generation = nilfs->ns_next_generation++;
394 spin_unlock(&nilfs->ns_next_gen_lock);
395 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
396 err = -EIO;
397 goto failed_after_creation;
398 }
399
400 err = nilfs_init_acl(inode, dir);
401 if (unlikely(err))
402 goto failed_after_creation; /* never occur. When supporting
403 nilfs_init_acl(), proper cancellation of
404 above jobs should be considered */
405
406 return inode;
407
408 failed_after_creation:
409 clear_nlink(inode);
410 unlock_new_inode(inode);
411 iput(inode); /* raw_inode will be deleted through
412 nilfs_evict_inode() */
413 goto failed;
414
415 failed_ifile_create_inode:
416 make_bad_inode(inode);
417 iput(inode); /* if i_nlink == 1, generic_forget_inode() will be
418 called */
419 failed:
420 return ERR_PTR(err);
421}
422
423void nilfs_set_inode_flags(struct inode *inode)
424{
425 unsigned int flags = NILFS_I(inode)->i_flags;
426 unsigned int new_fl = 0;
427
428 if (flags & FS_SYNC_FL)
429 new_fl |= S_SYNC;
430 if (flags & FS_APPEND_FL)
431 new_fl |= S_APPEND;
432 if (flags & FS_IMMUTABLE_FL)
433 new_fl |= S_IMMUTABLE;
434 if (flags & FS_NOATIME_FL)
435 new_fl |= S_NOATIME;
436 if (flags & FS_DIRSYNC_FL)
437 new_fl |= S_DIRSYNC;
438 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
439 S_NOATIME | S_DIRSYNC);
440}
441
442int nilfs_read_inode_common(struct inode *inode,
443 struct nilfs_inode *raw_inode)
444{
445 struct nilfs_inode_info *ii = NILFS_I(inode);
446 int err;
447
448 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
449 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
450 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
451 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
452 inode->i_size = le64_to_cpu(raw_inode->i_size);
453 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
454 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
455 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
456 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
457 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
458 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
459 if (inode->i_nlink == 0)
460 return -ESTALE; /* this inode is deleted */
461
462 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
463 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
464#if 0
465 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
466 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
467 0 : le32_to_cpu(raw_inode->i_dir_acl);
468#endif
469 ii->i_dir_start_lookup = 0;
470 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
471
472 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
473 S_ISLNK(inode->i_mode)) {
474 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
475 if (err < 0)
476 return err;
477 set_bit(NILFS_I_BMAP, &ii->i_state);
478 /* No lock is needed; iget() ensures it. */
479 }
480 return 0;
481}
482
483static int __nilfs_read_inode(struct super_block *sb,
484 struct nilfs_root *root, unsigned long ino,
485 struct inode *inode)
486{
487 struct the_nilfs *nilfs = sb->s_fs_info;
488 struct buffer_head *bh;
489 struct nilfs_inode *raw_inode;
490 int err;
491
492 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
493 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
494 if (unlikely(err))
495 goto bad_inode;
496
497 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
498
499 err = nilfs_read_inode_common(inode, raw_inode);
500 if (err)
501 goto failed_unmap;
502
503 if (S_ISREG(inode->i_mode)) {
504 inode->i_op = &nilfs_file_inode_operations;
505 inode->i_fop = &nilfs_file_operations;
506 inode->i_mapping->a_ops = &nilfs_aops;
507 } else if (S_ISDIR(inode->i_mode)) {
508 inode->i_op = &nilfs_dir_inode_operations;
509 inode->i_fop = &nilfs_dir_operations;
510 inode->i_mapping->a_ops = &nilfs_aops;
511 } else if (S_ISLNK(inode->i_mode)) {
512 inode->i_op = &nilfs_symlink_inode_operations;
513 inode_nohighmem(inode);
514 inode->i_mapping->a_ops = &nilfs_aops;
515 } else {
516 inode->i_op = &nilfs_special_inode_operations;
517 init_special_inode(
518 inode, inode->i_mode,
519 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
520 }
521 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
522 brelse(bh);
523 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
524 nilfs_set_inode_flags(inode);
525 mapping_set_gfp_mask(inode->i_mapping,
526 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
527 return 0;
528
529 failed_unmap:
530 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
531 brelse(bh);
532
533 bad_inode:
534 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
535 return err;
536}
537
538static int nilfs_iget_test(struct inode *inode, void *opaque)
539{
540 struct nilfs_iget_args *args = opaque;
541 struct nilfs_inode_info *ii;
542
543 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
544 return 0;
545
546 ii = NILFS_I(inode);
547 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
548 return !args->for_gc;
549
550 return args->for_gc && args->cno == ii->i_cno;
551}
552
553static int nilfs_iget_set(struct inode *inode, void *opaque)
554{
555 struct nilfs_iget_args *args = opaque;
556
557 inode->i_ino = args->ino;
558 if (args->for_gc) {
559 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
560 NILFS_I(inode)->i_cno = args->cno;
561 NILFS_I(inode)->i_root = NULL;
562 } else {
563 if (args->root && args->ino == NILFS_ROOT_INO)
564 nilfs_get_root(args->root);
565 NILFS_I(inode)->i_root = args->root;
566 }
567 return 0;
568}
569
570struct inode *nilfs_ilookup(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 ilookup5(sb, ino, nilfs_iget_test, &args);
578}
579
580struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
581 unsigned long ino)
582{
583 struct nilfs_iget_args args = {
584 .ino = ino, .root = root, .cno = 0, .for_gc = 0
585 };
586
587 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
588}
589
590struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
591 unsigned long ino)
592{
593 struct inode *inode;
594 int err;
595
596 inode = nilfs_iget_locked(sb, root, ino);
597 if (unlikely(!inode))
598 return ERR_PTR(-ENOMEM);
599 if (!(inode->i_state & I_NEW))
600 return inode;
601
602 err = __nilfs_read_inode(sb, root, ino, inode);
603 if (unlikely(err)) {
604 iget_failed(inode);
605 return ERR_PTR(err);
606 }
607 unlock_new_inode(inode);
608 return inode;
609}
610
611struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
612 __u64 cno)
613{
614 struct nilfs_iget_args args = {
615 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
616 };
617 struct inode *inode;
618 int err;
619
620 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
621 if (unlikely(!inode))
622 return ERR_PTR(-ENOMEM);
623 if (!(inode->i_state & I_NEW))
624 return inode;
625
626 err = nilfs_init_gcinode(inode);
627 if (unlikely(err)) {
628 iget_failed(inode);
629 return ERR_PTR(err);
630 }
631 unlock_new_inode(inode);
632 return inode;
633}
634
635void nilfs_write_inode_common(struct inode *inode,
636 struct nilfs_inode *raw_inode, int has_bmap)
637{
638 struct nilfs_inode_info *ii = NILFS_I(inode);
639
640 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
641 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
642 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
643 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
644 raw_inode->i_size = cpu_to_le64(inode->i_size);
645 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
646 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
647 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
648 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
649 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
650
651 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
652 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
653
654 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
655 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
656
657 /* zero-fill unused portion in the case of super root block */
658 raw_inode->i_xattr = 0;
659 raw_inode->i_pad = 0;
660 memset((void *)raw_inode + sizeof(*raw_inode), 0,
661 nilfs->ns_inode_size - sizeof(*raw_inode));
662 }
663
664 if (has_bmap)
665 nilfs_bmap_write(ii->i_bmap, raw_inode);
666 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
667 raw_inode->i_device_code =
668 cpu_to_le64(huge_encode_dev(inode->i_rdev));
669 /* When extending inode, nilfs->ns_inode_size should be checked
670 for substitutions of appended fields */
671}
672
673void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
674{
675 ino_t ino = inode->i_ino;
676 struct nilfs_inode_info *ii = NILFS_I(inode);
677 struct inode *ifile = ii->i_root->ifile;
678 struct nilfs_inode *raw_inode;
679
680 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
681
682 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
683 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
684 if (flags & I_DIRTY_DATASYNC)
685 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
686
687 nilfs_write_inode_common(inode, raw_inode, 0);
688 /* XXX: call with has_bmap = 0 is a workaround to avoid
689 deadlock of bmap. This delays update of i_bmap to just
690 before writing */
691 nilfs_ifile_unmap_inode(ifile, ino, ibh);
692}
693
694#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
695
696static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
697 unsigned long from)
698{
699 __u64 b;
700 int ret;
701
702 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
703 return;
704repeat:
705 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
706 if (ret == -ENOENT)
707 return;
708 else if (ret < 0)
709 goto failed;
710
711 if (b < from)
712 return;
713
714 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
715 ret = nilfs_bmap_truncate(ii->i_bmap, b);
716 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
717 if (!ret || (ret == -ENOMEM &&
718 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
719 goto repeat;
720
721failed:
722 nilfs_warning(ii->vfs_inode.i_sb, __func__,
723 "failed to truncate bmap (ino=%lu, err=%d)",
724 ii->vfs_inode.i_ino, ret);
725}
726
727void nilfs_truncate(struct inode *inode)
728{
729 unsigned long blkoff;
730 unsigned int blocksize;
731 struct nilfs_transaction_info ti;
732 struct super_block *sb = inode->i_sb;
733 struct nilfs_inode_info *ii = NILFS_I(inode);
734
735 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
736 return;
737 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
738 return;
739
740 blocksize = sb->s_blocksize;
741 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
742 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
743
744 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
745
746 nilfs_truncate_bmap(ii, blkoff);
747
748 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
749 if (IS_SYNC(inode))
750 nilfs_set_transaction_flag(NILFS_TI_SYNC);
751
752 nilfs_mark_inode_dirty(inode);
753 nilfs_set_file_dirty(inode, 0);
754 nilfs_transaction_commit(sb);
755 /* May construct a logical segment and may fail in sync mode.
756 But truncate has no return value. */
757}
758
759static void nilfs_clear_inode(struct inode *inode)
760{
761 struct nilfs_inode_info *ii = NILFS_I(inode);
762 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
763
764 /*
765 * Free resources allocated in nilfs_read_inode(), here.
766 */
767 BUG_ON(!list_empty(&ii->i_dirty));
768 brelse(ii->i_bh);
769 ii->i_bh = NULL;
770
771 if (mdi && mdi->mi_palloc_cache)
772 nilfs_palloc_destroy_cache(inode);
773
774 if (test_bit(NILFS_I_BMAP, &ii->i_state))
775 nilfs_bmap_clear(ii->i_bmap);
776
777 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
778
779 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
780 nilfs_put_root(ii->i_root);
781}
782
783void nilfs_evict_inode(struct inode *inode)
784{
785 struct nilfs_transaction_info ti;
786 struct super_block *sb = inode->i_sb;
787 struct nilfs_inode_info *ii = NILFS_I(inode);
788 int ret;
789
790 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
791 truncate_inode_pages_final(&inode->i_data);
792 clear_inode(inode);
793 nilfs_clear_inode(inode);
794 return;
795 }
796 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
797
798 truncate_inode_pages_final(&inode->i_data);
799
800 /* TODO: some of the following operations may fail. */
801 nilfs_truncate_bmap(ii, 0);
802 nilfs_mark_inode_dirty(inode);
803 clear_inode(inode);
804
805 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
806 if (!ret)
807 atomic64_dec(&ii->i_root->inodes_count);
808
809 nilfs_clear_inode(inode);
810
811 if (IS_SYNC(inode))
812 nilfs_set_transaction_flag(NILFS_TI_SYNC);
813 nilfs_transaction_commit(sb);
814 /* May construct a logical segment and may fail in sync mode.
815 But delete_inode has no return value. */
816}
817
818int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
819{
820 struct nilfs_transaction_info ti;
821 struct inode *inode = d_inode(dentry);
822 struct super_block *sb = inode->i_sb;
823 int err;
824
825 err = inode_change_ok(inode, iattr);
826 if (err)
827 return err;
828
829 err = nilfs_transaction_begin(sb, &ti, 0);
830 if (unlikely(err))
831 return err;
832
833 if ((iattr->ia_valid & ATTR_SIZE) &&
834 iattr->ia_size != i_size_read(inode)) {
835 inode_dio_wait(inode);
836 truncate_setsize(inode, iattr->ia_size);
837 nilfs_truncate(inode);
838 }
839
840 setattr_copy(inode, iattr);
841 mark_inode_dirty(inode);
842
843 if (iattr->ia_valid & ATTR_MODE) {
844 err = nilfs_acl_chmod(inode);
845 if (unlikely(err))
846 goto out_err;
847 }
848
849 return nilfs_transaction_commit(sb);
850
851out_err:
852 nilfs_transaction_abort(sb);
853 return err;
854}
855
856int nilfs_permission(struct inode *inode, int mask)
857{
858 struct nilfs_root *root = NILFS_I(inode)->i_root;
859 if ((mask & MAY_WRITE) && root &&
860 root->cno != NILFS_CPTREE_CURRENT_CNO)
861 return -EROFS; /* snapshot is not writable */
862
863 return generic_permission(inode, mask);
864}
865
866int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
867{
868 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
869 struct nilfs_inode_info *ii = NILFS_I(inode);
870 int err;
871
872 spin_lock(&nilfs->ns_inode_lock);
873 if (ii->i_bh == NULL) {
874 spin_unlock(&nilfs->ns_inode_lock);
875 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
876 inode->i_ino, pbh);
877 if (unlikely(err))
878 return err;
879 spin_lock(&nilfs->ns_inode_lock);
880 if (ii->i_bh == NULL)
881 ii->i_bh = *pbh;
882 else {
883 brelse(*pbh);
884 *pbh = ii->i_bh;
885 }
886 } else
887 *pbh = ii->i_bh;
888
889 get_bh(*pbh);
890 spin_unlock(&nilfs->ns_inode_lock);
891 return 0;
892}
893
894int nilfs_inode_dirty(struct inode *inode)
895{
896 struct nilfs_inode_info *ii = NILFS_I(inode);
897 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
898 int ret = 0;
899
900 if (!list_empty(&ii->i_dirty)) {
901 spin_lock(&nilfs->ns_inode_lock);
902 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
903 test_bit(NILFS_I_BUSY, &ii->i_state);
904 spin_unlock(&nilfs->ns_inode_lock);
905 }
906 return ret;
907}
908
909int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
910{
911 struct nilfs_inode_info *ii = NILFS_I(inode);
912 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
913
914 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
915
916 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
917 return 0;
918
919 spin_lock(&nilfs->ns_inode_lock);
920 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
921 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
922 /* Because this routine may race with nilfs_dispose_list(),
923 we have to check NILFS_I_QUEUED here, too. */
924 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
925 /* This will happen when somebody is freeing
926 this inode. */
927 nilfs_warning(inode->i_sb, __func__,
928 "cannot get inode (ino=%lu)\n",
929 inode->i_ino);
930 spin_unlock(&nilfs->ns_inode_lock);
931 return -EINVAL; /* NILFS_I_DIRTY may remain for
932 freeing inode */
933 }
934 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
935 set_bit(NILFS_I_QUEUED, &ii->i_state);
936 }
937 spin_unlock(&nilfs->ns_inode_lock);
938 return 0;
939}
940
941int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
942{
943 struct buffer_head *ibh;
944 int err;
945
946 err = nilfs_load_inode_block(inode, &ibh);
947 if (unlikely(err)) {
948 nilfs_warning(inode->i_sb, __func__,
949 "failed to reget inode block.\n");
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_warning(inode->i_sb, __func__,
976 "tried to mark bad_inode dirty. ignored.\n");
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_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
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}
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}