1 // SPDX-License-Identifier: GPL-2.0+
3 * NILFS inode operations.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
11 #include <linux/buffer_head.h>
12 #include <linux/gfp.h>
13 #include <linux/mpage.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/uio.h>
17 #include <linux/fiemap.h>
27 * struct nilfs_iget_args - arguments used during comparison between inodes
29 * @cno: checkpoint number
30 * @root: pointer on NILFS root object (mounted checkpoint)
31 * @for_gc: inode for GC flag
32 * @for_btnc: inode for B-tree node cache flag
33 * @for_shadow: inode for shadowed page cache flag
35 struct nilfs_iget_args {
38 struct nilfs_root *root;
44 static int nilfs_iget_test(struct inode *inode, void *opaque);
46 void nilfs_inode_add_blocks(struct inode *inode, int n)
48 struct nilfs_root *root = NILFS_I(inode)->i_root;
50 inode_add_bytes(inode, i_blocksize(inode) * n);
52 atomic64_add(n, &root->blocks_count);
55 void nilfs_inode_sub_blocks(struct inode *inode, int n)
57 struct nilfs_root *root = NILFS_I(inode)->i_root;
59 inode_sub_bytes(inode, i_blocksize(inode) * n);
61 atomic64_sub(n, &root->blocks_count);
65 * nilfs_get_block() - get a file block on the filesystem (callback function)
66 * @inode: inode struct of the target file
67 * @blkoff: file block number
68 * @bh_result: buffer head to be mapped on
69 * @create: indicate whether allocating the block or not when it has not
72 * This function does not issue actual read request of the specified data
73 * block. It is done by VFS.
75 int nilfs_get_block(struct inode *inode, sector_t blkoff,
76 struct buffer_head *bh_result, int create)
78 struct nilfs_inode_info *ii = NILFS_I(inode);
79 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
82 unsigned int maxblocks = bh_result->b_size >> inode->i_blkbits;
84 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
85 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
86 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
87 if (ret >= 0) { /* found */
88 map_bh(bh_result, inode->i_sb, blknum);
90 bh_result->b_size = (ret << inode->i_blkbits);
93 /* data block was not found */
94 if (ret == -ENOENT && create) {
95 struct nilfs_transaction_info ti;
97 bh_result->b_blocknr = 0;
98 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
101 err = nilfs_bmap_insert(ii->i_bmap, blkoff,
102 (unsigned long)bh_result);
103 if (unlikely(err != 0)) {
104 if (err == -EEXIST) {
106 * The get_block() function could be called
107 * from multiple callers for an inode.
108 * However, the page having this block must
109 * be locked in this case.
111 nilfs_warn(inode->i_sb,
112 "%s (ino=%lu): a race condition while inserting a data block at offset=%llu",
113 __func__, inode->i_ino,
114 (unsigned long long)blkoff);
117 nilfs_transaction_abort(inode->i_sb);
120 nilfs_mark_inode_dirty_sync(inode);
121 nilfs_transaction_commit(inode->i_sb); /* never fails */
122 /* Error handling should be detailed */
123 set_buffer_new(bh_result);
124 set_buffer_delay(bh_result);
125 map_bh(bh_result, inode->i_sb, 0);
126 /* Disk block number must be changed to proper value */
128 } else if (ret == -ENOENT) {
130 * not found is not error (e.g. hole); must return without
131 * the mapped state flag.
143 * nilfs_read_folio() - implement read_folio() method of nilfs_aops {}
144 * address_space_operations.
145 * @file: file struct of the file to be read
146 * @folio: the folio to be read
148 static int nilfs_read_folio(struct file *file, struct folio *folio)
150 return mpage_read_folio(folio, nilfs_get_block);
153 static void nilfs_readahead(struct readahead_control *rac)
155 mpage_readahead(rac, nilfs_get_block);
158 static int nilfs_writepages(struct address_space *mapping,
159 struct writeback_control *wbc)
161 struct inode *inode = mapping->host;
164 if (sb_rdonly(inode->i_sb)) {
165 nilfs_clear_dirty_pages(mapping, false);
169 if (wbc->sync_mode == WB_SYNC_ALL)
170 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
176 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
178 struct folio *folio = page_folio(page);
179 struct inode *inode = folio->mapping->host;
182 if (sb_rdonly(inode->i_sb)) {
184 * It means that filesystem was remounted in read-only
185 * mode because of error or metadata corruption. But we
186 * have dirty pages that try to be flushed in background.
187 * So, here we simply discard this dirty page.
189 nilfs_clear_folio_dirty(folio, false);
194 folio_redirty_for_writepage(wbc, folio);
197 if (wbc->sync_mode == WB_SYNC_ALL) {
198 err = nilfs_construct_segment(inode->i_sb);
201 } else if (wbc->for_reclaim)
202 nilfs_flush_segment(inode->i_sb, inode->i_ino);
207 static bool nilfs_dirty_folio(struct address_space *mapping,
210 struct inode *inode = mapping->host;
211 struct buffer_head *head;
212 unsigned int nr_dirty = 0;
213 bool ret = filemap_dirty_folio(mapping, folio);
216 * The page may not be locked, eg if called from try_to_unmap_one()
218 spin_lock(&mapping->i_private_lock);
219 head = folio_buffers(folio);
221 struct buffer_head *bh = head;
224 /* Do not mark hole blocks dirty */
225 if (buffer_dirty(bh) || !buffer_mapped(bh))
228 set_buffer_dirty(bh);
230 } while (bh = bh->b_this_page, bh != head);
232 nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
234 spin_unlock(&mapping->i_private_lock);
237 nilfs_set_file_dirty(inode, nr_dirty);
241 void nilfs_write_failed(struct address_space *mapping, loff_t to)
243 struct inode *inode = mapping->host;
245 if (to > inode->i_size) {
246 truncate_pagecache(inode, inode->i_size);
247 nilfs_truncate(inode);
251 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
252 loff_t pos, unsigned len,
253 struct page **pagep, void **fsdata)
256 struct inode *inode = mapping->host;
257 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
262 err = block_write_begin(mapping, pos, len, pagep, nilfs_get_block);
264 nilfs_write_failed(mapping, pos + len);
265 nilfs_transaction_abort(inode->i_sb);
270 static int nilfs_write_end(struct file *file, struct address_space *mapping,
271 loff_t pos, unsigned len, unsigned copied,
272 struct page *page, void *fsdata)
274 struct inode *inode = mapping->host;
275 unsigned int start = pos & (PAGE_SIZE - 1);
276 unsigned int nr_dirty;
279 nr_dirty = nilfs_page_count_clean_buffers(page, start,
281 copied = generic_write_end(file, mapping, pos, len, copied, page,
283 nilfs_set_file_dirty(inode, nr_dirty);
284 err = nilfs_transaction_commit(inode->i_sb);
285 return err ? : copied;
289 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
291 struct inode *inode = file_inode(iocb->ki_filp);
293 if (iov_iter_rw(iter) == WRITE)
296 /* Needs synchronization with the cleaner */
297 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
300 const struct address_space_operations nilfs_aops = {
301 .writepage = nilfs_writepage,
302 .read_folio = nilfs_read_folio,
303 .writepages = nilfs_writepages,
304 .dirty_folio = nilfs_dirty_folio,
305 .readahead = nilfs_readahead,
306 .write_begin = nilfs_write_begin,
307 .write_end = nilfs_write_end,
308 .invalidate_folio = block_invalidate_folio,
309 .direct_IO = nilfs_direct_IO,
310 .is_partially_uptodate = block_is_partially_uptodate,
313 static int nilfs_insert_inode_locked(struct inode *inode,
314 struct nilfs_root *root,
317 struct nilfs_iget_args args = {
318 .ino = ino, .root = root, .cno = 0, .for_gc = false,
319 .for_btnc = false, .for_shadow = false
322 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
325 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
327 struct super_block *sb = dir->i_sb;
328 struct the_nilfs *nilfs = sb->s_fs_info;
330 struct nilfs_inode_info *ii;
331 struct nilfs_root *root;
332 struct buffer_head *bh;
336 inode = new_inode(sb);
337 if (unlikely(!inode))
340 mapping_set_gfp_mask(inode->i_mapping,
341 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
343 root = NILFS_I(dir)->i_root;
345 ii->i_state = BIT(NILFS_I_NEW);
348 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
350 goto failed_ifile_create_inode;
351 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
353 if (unlikely(ino < NILFS_USER_INO)) {
355 "inode bitmap is inconsistent for reserved inodes");
358 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
360 goto failed_ifile_create_inode;
361 } while (ino < NILFS_USER_INO);
363 nilfs_info(sb, "repaired inode bitmap for reserved inodes");
367 atomic64_inc(&root->inodes_count);
368 inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
370 simple_inode_init_ts(inode);
372 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
373 err = nilfs_bmap_read(ii->i_bmap, NULL);
375 goto failed_after_creation;
377 set_bit(NILFS_I_BMAP, &ii->i_state);
378 /* No lock is needed; iget() ensures it. */
381 ii->i_flags = nilfs_mask_flags(
382 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
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) {
393 goto failed_after_creation;
396 err = nilfs_init_acl(inode, dir);
399 * Never occur. When supporting nilfs_init_acl(),
400 * proper cancellation of above jobs should be considered.
402 goto failed_after_creation;
406 failed_after_creation:
408 if (inode->i_state & I_NEW)
409 unlock_new_inode(inode);
411 * raw_inode will be deleted through
412 * nilfs_evict_inode().
416 failed_ifile_create_inode:
417 make_bad_inode(inode);
423 void nilfs_set_inode_flags(struct inode *inode)
425 unsigned int flags = NILFS_I(inode)->i_flags;
426 unsigned int new_fl = 0;
428 if (flags & FS_SYNC_FL)
430 if (flags & FS_APPEND_FL)
432 if (flags & FS_IMMUTABLE_FL)
433 new_fl |= S_IMMUTABLE;
434 if (flags & FS_NOATIME_FL)
436 if (flags & FS_DIRSYNC_FL)
438 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
439 S_NOATIME | S_DIRSYNC);
442 int nilfs_read_inode_common(struct inode *inode,
443 struct nilfs_inode *raw_inode)
445 struct nilfs_inode_info *ii = NILFS_I(inode);
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_set_atime(inode, le64_to_cpu(raw_inode->i_mtime),
454 le32_to_cpu(raw_inode->i_mtime_nsec));
455 inode_set_ctime(inode, le64_to_cpu(raw_inode->i_ctime),
456 le32_to_cpu(raw_inode->i_ctime_nsec));
457 inode_set_mtime(inode, le64_to_cpu(raw_inode->i_mtime),
458 le32_to_cpu(raw_inode->i_mtime_nsec));
459 if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
460 return -EIO; /* this inode is for metadata and corrupted */
461 if (inode->i_nlink == 0)
462 return -ESTALE; /* this inode is deleted */
464 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
465 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
467 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
468 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
469 0 : le32_to_cpu(raw_inode->i_dir_acl);
471 ii->i_dir_start_lookup = 0;
472 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
474 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
475 S_ISLNK(inode->i_mode)) {
476 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
479 set_bit(NILFS_I_BMAP, &ii->i_state);
480 /* No lock is needed; iget() ensures it. */
485 static int __nilfs_read_inode(struct super_block *sb,
486 struct nilfs_root *root, unsigned long ino,
489 struct the_nilfs *nilfs = sb->s_fs_info;
490 struct buffer_head *bh;
491 struct nilfs_inode *raw_inode;
494 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
495 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
499 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
501 err = nilfs_read_inode_common(inode, raw_inode);
505 if (S_ISREG(inode->i_mode)) {
506 inode->i_op = &nilfs_file_inode_operations;
507 inode->i_fop = &nilfs_file_operations;
508 inode->i_mapping->a_ops = &nilfs_aops;
509 } else if (S_ISDIR(inode->i_mode)) {
510 inode->i_op = &nilfs_dir_inode_operations;
511 inode->i_fop = &nilfs_dir_operations;
512 inode->i_mapping->a_ops = &nilfs_aops;
513 } else if (S_ISLNK(inode->i_mode)) {
514 inode->i_op = &nilfs_symlink_inode_operations;
515 inode_nohighmem(inode);
516 inode->i_mapping->a_ops = &nilfs_aops;
518 inode->i_op = &nilfs_special_inode_operations;
520 inode, inode->i_mode,
521 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
523 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
525 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
526 nilfs_set_inode_flags(inode);
527 mapping_set_gfp_mask(inode->i_mapping,
528 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
532 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
536 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
540 static int nilfs_iget_test(struct inode *inode, void *opaque)
542 struct nilfs_iget_args *args = opaque;
543 struct nilfs_inode_info *ii;
545 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
549 if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
552 } else if (args->for_btnc) {
555 if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
556 if (!args->for_shadow)
558 } else if (args->for_shadow) {
562 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
563 return !args->for_gc;
565 return args->for_gc && args->cno == ii->i_cno;
568 static int nilfs_iget_set(struct inode *inode, void *opaque)
570 struct nilfs_iget_args *args = opaque;
572 inode->i_ino = args->ino;
573 NILFS_I(inode)->i_cno = args->cno;
574 NILFS_I(inode)->i_root = args->root;
575 if (args->root && args->ino == NILFS_ROOT_INO)
576 nilfs_get_root(args->root);
579 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
581 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
582 if (args->for_shadow)
583 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
587 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
590 struct nilfs_iget_args args = {
591 .ino = ino, .root = root, .cno = 0, .for_gc = false,
592 .for_btnc = false, .for_shadow = false
595 return ilookup5(sb, ino, nilfs_iget_test, &args);
598 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
601 struct nilfs_iget_args args = {
602 .ino = ino, .root = root, .cno = 0, .for_gc = false,
603 .for_btnc = false, .for_shadow = false
606 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
609 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
615 inode = nilfs_iget_locked(sb, root, ino);
616 if (unlikely(!inode))
617 return ERR_PTR(-ENOMEM);
618 if (!(inode->i_state & I_NEW))
621 err = __nilfs_read_inode(sb, root, ino, inode);
626 unlock_new_inode(inode);
630 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
633 struct nilfs_iget_args args = {
634 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
635 .for_btnc = false, .for_shadow = false
640 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
641 if (unlikely(!inode))
642 return ERR_PTR(-ENOMEM);
643 if (!(inode->i_state & I_NEW))
646 err = nilfs_init_gcinode(inode);
651 unlock_new_inode(inode);
656 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
657 * @inode: inode object
659 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
660 * or does nothing if the inode already has it. This function allocates
661 * an additional inode to maintain page cache of B-tree nodes one-on-one.
663 * Return Value: On success, 0 is returned. On errors, one of the following
664 * negative error code is returned.
666 * %-ENOMEM - Insufficient memory available.
668 int nilfs_attach_btree_node_cache(struct inode *inode)
670 struct nilfs_inode_info *ii = NILFS_I(inode);
671 struct inode *btnc_inode;
672 struct nilfs_iget_args args;
674 if (ii->i_assoc_inode)
677 args.ino = inode->i_ino;
678 args.root = ii->i_root;
679 args.cno = ii->i_cno;
680 args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
681 args.for_btnc = true;
682 args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
684 btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
685 nilfs_iget_set, &args);
686 if (unlikely(!btnc_inode))
688 if (btnc_inode->i_state & I_NEW) {
689 nilfs_init_btnc_inode(btnc_inode);
690 unlock_new_inode(btnc_inode);
692 NILFS_I(btnc_inode)->i_assoc_inode = inode;
693 NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
694 ii->i_assoc_inode = btnc_inode;
700 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
701 * @inode: inode object
703 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
704 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
706 void nilfs_detach_btree_node_cache(struct inode *inode)
708 struct nilfs_inode_info *ii = NILFS_I(inode);
709 struct inode *btnc_inode = ii->i_assoc_inode;
712 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
713 ii->i_assoc_inode = NULL;
719 * nilfs_iget_for_shadow - obtain inode for shadow mapping
720 * @inode: inode object that uses shadow mapping
722 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
723 * caches for shadow mapping. The page cache for data pages is set up
724 * in one inode and the one for b-tree node pages is set up in the
725 * other inode, which is attached to the former inode.
727 * Return Value: On success, a pointer to the inode for data pages is
728 * returned. On errors, one of the following negative error code is returned
731 * %-ENOMEM - Insufficient memory available.
733 struct inode *nilfs_iget_for_shadow(struct inode *inode)
735 struct nilfs_iget_args args = {
736 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
737 .for_btnc = false, .for_shadow = true
739 struct inode *s_inode;
742 s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
743 nilfs_iget_set, &args);
744 if (unlikely(!s_inode))
745 return ERR_PTR(-ENOMEM);
746 if (!(s_inode->i_state & I_NEW))
749 NILFS_I(s_inode)->i_flags = 0;
750 memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
751 mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
753 err = nilfs_attach_btree_node_cache(s_inode);
755 iget_failed(s_inode);
758 unlock_new_inode(s_inode);
762 void nilfs_write_inode_common(struct inode *inode,
763 struct nilfs_inode *raw_inode, int has_bmap)
765 struct nilfs_inode_info *ii = NILFS_I(inode);
767 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
768 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
769 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
770 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
771 raw_inode->i_size = cpu_to_le64(inode->i_size);
772 raw_inode->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
773 raw_inode->i_mtime = cpu_to_le64(inode_get_mtime_sec(inode));
774 raw_inode->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
775 raw_inode->i_mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
776 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
778 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
779 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
781 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
782 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
784 /* zero-fill unused portion in the case of super root block */
785 raw_inode->i_xattr = 0;
786 raw_inode->i_pad = 0;
787 memset((void *)raw_inode + sizeof(*raw_inode), 0,
788 nilfs->ns_inode_size - sizeof(*raw_inode));
792 nilfs_bmap_write(ii->i_bmap, raw_inode);
793 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
794 raw_inode->i_device_code =
795 cpu_to_le64(huge_encode_dev(inode->i_rdev));
797 * When extending inode, nilfs->ns_inode_size should be checked
798 * for substitutions of appended fields.
802 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
804 ino_t ino = inode->i_ino;
805 struct nilfs_inode_info *ii = NILFS_I(inode);
806 struct inode *ifile = ii->i_root->ifile;
807 struct nilfs_inode *raw_inode;
809 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
811 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
812 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
813 if (flags & I_DIRTY_DATASYNC)
814 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
816 nilfs_write_inode_common(inode, raw_inode, 0);
818 * XXX: call with has_bmap = 0 is a workaround to avoid
819 * deadlock of bmap. This delays update of i_bmap to just
823 nilfs_ifile_unmap_inode(ifile, ino, ibh);
826 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
828 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
834 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
837 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
846 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
847 ret = nilfs_bmap_truncate(ii->i_bmap, b);
848 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
849 if (!ret || (ret == -ENOMEM &&
850 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
854 nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
855 ret, ii->vfs_inode.i_ino);
858 void nilfs_truncate(struct inode *inode)
860 unsigned long blkoff;
861 unsigned int blocksize;
862 struct nilfs_transaction_info ti;
863 struct super_block *sb = inode->i_sb;
864 struct nilfs_inode_info *ii = NILFS_I(inode);
866 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
868 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
871 blocksize = sb->s_blocksize;
872 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
873 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
875 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
877 nilfs_truncate_bmap(ii, blkoff);
879 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
881 nilfs_set_transaction_flag(NILFS_TI_SYNC);
883 nilfs_mark_inode_dirty(inode);
884 nilfs_set_file_dirty(inode, 0);
885 nilfs_transaction_commit(sb);
887 * May construct a logical segment and may fail in sync mode.
888 * But truncate has no return value.
892 static void nilfs_clear_inode(struct inode *inode)
894 struct nilfs_inode_info *ii = NILFS_I(inode);
897 * Free resources allocated in nilfs_read_inode(), here.
899 BUG_ON(!list_empty(&ii->i_dirty));
903 if (nilfs_is_metadata_file_inode(inode))
904 nilfs_mdt_clear(inode);
906 if (test_bit(NILFS_I_BMAP, &ii->i_state))
907 nilfs_bmap_clear(ii->i_bmap);
909 if (!test_bit(NILFS_I_BTNC, &ii->i_state))
910 nilfs_detach_btree_node_cache(inode);
912 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
913 nilfs_put_root(ii->i_root);
916 void nilfs_evict_inode(struct inode *inode)
918 struct nilfs_transaction_info ti;
919 struct super_block *sb = inode->i_sb;
920 struct nilfs_inode_info *ii = NILFS_I(inode);
921 struct the_nilfs *nilfs;
924 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
925 truncate_inode_pages_final(&inode->i_data);
927 nilfs_clear_inode(inode);
930 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
932 truncate_inode_pages_final(&inode->i_data);
934 nilfs = sb->s_fs_info;
935 if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
937 * If this inode is about to be disposed after the file system
938 * has been degraded to read-only due to file system corruption
939 * or after the writer has been detached, do not make any
940 * changes that cause writes, just clear it.
941 * Do this check after read-locking ns_segctor_sem by
942 * nilfs_transaction_begin() in order to avoid a race with
943 * the writer detach operation.
946 nilfs_clear_inode(inode);
947 nilfs_transaction_abort(sb);
951 /* TODO: some of the following operations may fail. */
952 nilfs_truncate_bmap(ii, 0);
953 nilfs_mark_inode_dirty(inode);
956 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
958 atomic64_dec(&ii->i_root->inodes_count);
960 nilfs_clear_inode(inode);
963 nilfs_set_transaction_flag(NILFS_TI_SYNC);
964 nilfs_transaction_commit(sb);
966 * May construct a logical segment and may fail in sync mode.
967 * But delete_inode has no return value.
971 int nilfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
974 struct nilfs_transaction_info ti;
975 struct inode *inode = d_inode(dentry);
976 struct super_block *sb = inode->i_sb;
979 err = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
983 err = nilfs_transaction_begin(sb, &ti, 0);
987 if ((iattr->ia_valid & ATTR_SIZE) &&
988 iattr->ia_size != i_size_read(inode)) {
989 inode_dio_wait(inode);
990 truncate_setsize(inode, iattr->ia_size);
991 nilfs_truncate(inode);
994 setattr_copy(&nop_mnt_idmap, inode, iattr);
995 mark_inode_dirty(inode);
997 if (iattr->ia_valid & ATTR_MODE) {
998 err = nilfs_acl_chmod(inode);
1003 return nilfs_transaction_commit(sb);
1006 nilfs_transaction_abort(sb);
1010 int nilfs_permission(struct mnt_idmap *idmap, struct inode *inode,
1013 struct nilfs_root *root = NILFS_I(inode)->i_root;
1015 if ((mask & MAY_WRITE) && root &&
1016 root->cno != NILFS_CPTREE_CURRENT_CNO)
1017 return -EROFS; /* snapshot is not writable */
1019 return generic_permission(&nop_mnt_idmap, inode, mask);
1022 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1024 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1025 struct nilfs_inode_info *ii = NILFS_I(inode);
1028 spin_lock(&nilfs->ns_inode_lock);
1029 if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1030 spin_unlock(&nilfs->ns_inode_lock);
1031 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1035 spin_lock(&nilfs->ns_inode_lock);
1036 if (ii->i_bh == NULL)
1038 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1049 spin_unlock(&nilfs->ns_inode_lock);
1053 int nilfs_inode_dirty(struct inode *inode)
1055 struct nilfs_inode_info *ii = NILFS_I(inode);
1056 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1059 if (!list_empty(&ii->i_dirty)) {
1060 spin_lock(&nilfs->ns_inode_lock);
1061 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1062 test_bit(NILFS_I_BUSY, &ii->i_state);
1063 spin_unlock(&nilfs->ns_inode_lock);
1068 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1070 struct nilfs_inode_info *ii = NILFS_I(inode);
1071 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1073 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1075 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1078 spin_lock(&nilfs->ns_inode_lock);
1079 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1080 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1082 * Because this routine may race with nilfs_dispose_list(),
1083 * we have to check NILFS_I_QUEUED here, too.
1085 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1087 * This will happen when somebody is freeing
1090 nilfs_warn(inode->i_sb,
1091 "cannot set file dirty (ino=%lu): the file is being freed",
1093 spin_unlock(&nilfs->ns_inode_lock);
1095 * NILFS_I_DIRTY may remain for
1099 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1100 set_bit(NILFS_I_QUEUED, &ii->i_state);
1102 spin_unlock(&nilfs->ns_inode_lock);
1106 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1108 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1109 struct buffer_head *ibh;
1113 * Do not dirty inodes after the log writer has been detached
1114 * and its nilfs_root struct has been freed.
1116 if (unlikely(nilfs_purging(nilfs)))
1119 err = nilfs_load_inode_block(inode, &ibh);
1120 if (unlikely(err)) {
1121 nilfs_warn(inode->i_sb,
1122 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1126 nilfs_update_inode(inode, ibh, flags);
1127 mark_buffer_dirty(ibh);
1128 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1134 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1135 * @inode: inode of the file to be registered.
1136 * @flags: flags to determine the dirty state of the inode
1138 * nilfs_dirty_inode() loads a inode block containing the specified
1139 * @inode and copies data from a nilfs_inode to a corresponding inode
1140 * entry in the inode block. This operation is excluded from the segment
1141 * construction. This function can be called both as a single operation
1142 * and as a part of indivisible file operations.
1144 void nilfs_dirty_inode(struct inode *inode, int flags)
1146 struct nilfs_transaction_info ti;
1147 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1149 if (is_bad_inode(inode)) {
1150 nilfs_warn(inode->i_sb,
1151 "tried to mark bad_inode dirty. ignored.");
1156 nilfs_mdt_mark_dirty(inode);
1159 nilfs_transaction_begin(inode->i_sb, &ti, 0);
1160 __nilfs_mark_inode_dirty(inode, flags);
1161 nilfs_transaction_commit(inode->i_sb); /* never fails */
1164 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1165 __u64 start, __u64 len)
1167 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1168 __u64 logical = 0, phys = 0, size = 0;
1171 sector_t blkoff, end_blkoff;
1172 sector_t delalloc_blkoff;
1173 unsigned long delalloc_blklen;
1174 unsigned int blkbits = inode->i_blkbits;
1177 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1183 isize = i_size_read(inode);
1185 blkoff = start >> blkbits;
1186 end_blkoff = (start + len - 1) >> blkbits;
1188 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1193 unsigned int maxblocks;
1195 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1197 /* End of the current extent */
1198 ret = fiemap_fill_next_extent(
1199 fieinfo, logical, phys, size, flags);
1203 if (blkoff > end_blkoff)
1206 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1207 logical = blkoff << blkbits;
1209 size = delalloc_blklen << blkbits;
1211 blkoff = delalloc_blkoff + delalloc_blklen;
1212 delalloc_blklen = nilfs_find_uncommitted_extent(
1213 inode, blkoff, &delalloc_blkoff);
1218 * Limit the number of blocks that we look up so as
1219 * not to get into the next delayed allocation extent.
1221 maxblocks = INT_MAX;
1222 if (delalloc_blklen)
1223 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1227 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1228 n = nilfs_bmap_lookup_contig(
1229 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1230 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1235 if (unlikely(n != -ENOENT))
1240 past_eof = ((blkoff << blkbits) >= isize);
1243 /* End of the current extent */
1246 flags |= FIEMAP_EXTENT_LAST;
1248 ret = fiemap_fill_next_extent(
1249 fieinfo, logical, phys, size, flags);
1254 if (blkoff > end_blkoff || past_eof)
1258 if (phys && blkphy << blkbits == phys + size) {
1259 /* The current extent goes on */
1260 size += n << blkbits;
1262 /* Terminate the current extent */
1263 ret = fiemap_fill_next_extent(
1264 fieinfo, logical, phys, size,
1266 if (ret || blkoff > end_blkoff)
1269 /* Start another extent */
1270 flags = FIEMAP_EXTENT_MERGED;
1271 logical = blkoff << blkbits;
1272 phys = blkphy << blkbits;
1273 size = n << blkbits;
1276 /* Start a new extent */
1277 flags = FIEMAP_EXTENT_MERGED;
1278 logical = blkoff << blkbits;
1279 phys = blkphy << blkbits;
1280 size = n << blkbits;
1287 /* If ret is 1 then we just hit the end of the extent array */
1291 inode_unlock(inode);