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 inode *inode = page->mapping->host;
181 if (sb_rdonly(inode->i_sb)) {
183 * It means that filesystem was remounted in read-only
184 * mode because of error or metadata corruption. But we
185 * have dirty pages that try to be flushed in background.
186 * So, here we simply discard this dirty page.
188 nilfs_clear_dirty_page(page, false);
193 redirty_page_for_writepage(wbc, page);
196 if (wbc->sync_mode == WB_SYNC_ALL) {
197 err = nilfs_construct_segment(inode->i_sb);
200 } else if (wbc->for_reclaim)
201 nilfs_flush_segment(inode->i_sb, inode->i_ino);
206 static bool nilfs_dirty_folio(struct address_space *mapping,
209 struct inode *inode = mapping->host;
210 struct buffer_head *head;
211 unsigned int nr_dirty = 0;
212 bool ret = filemap_dirty_folio(mapping, folio);
215 * The page may not be locked, eg if called from try_to_unmap_one()
217 spin_lock(&mapping->private_lock);
218 head = folio_buffers(folio);
220 struct buffer_head *bh = head;
223 /* Do not mark hole blocks dirty */
224 if (buffer_dirty(bh) || !buffer_mapped(bh))
227 set_buffer_dirty(bh);
229 } while (bh = bh->b_this_page, bh != head);
231 nr_dirty = 1 << (folio_shift(folio) - inode->i_blkbits);
233 spin_unlock(&mapping->private_lock);
236 nilfs_set_file_dirty(inode, nr_dirty);
240 void nilfs_write_failed(struct address_space *mapping, loff_t to)
242 struct inode *inode = mapping->host;
244 if (to > inode->i_size) {
245 truncate_pagecache(inode, inode->i_size);
246 nilfs_truncate(inode);
250 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
251 loff_t pos, unsigned len,
252 struct page **pagep, void **fsdata)
255 struct inode *inode = mapping->host;
256 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
261 err = block_write_begin(mapping, pos, len, pagep, nilfs_get_block);
263 nilfs_write_failed(mapping, pos + len);
264 nilfs_transaction_abort(inode->i_sb);
269 static int nilfs_write_end(struct file *file, struct address_space *mapping,
270 loff_t pos, unsigned len, unsigned copied,
271 struct page *page, void *fsdata)
273 struct inode *inode = mapping->host;
274 unsigned int start = pos & (PAGE_SIZE - 1);
275 unsigned int nr_dirty;
278 nr_dirty = nilfs_page_count_clean_buffers(page, start,
280 copied = generic_write_end(file, mapping, pos, len, copied, page,
282 nilfs_set_file_dirty(inode, nr_dirty);
283 err = nilfs_transaction_commit(inode->i_sb);
284 return err ? : copied;
288 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
290 struct inode *inode = file_inode(iocb->ki_filp);
292 if (iov_iter_rw(iter) == WRITE)
295 /* Needs synchronization with the cleaner */
296 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
299 const struct address_space_operations nilfs_aops = {
300 .writepage = nilfs_writepage,
301 .read_folio = nilfs_read_folio,
302 .writepages = nilfs_writepages,
303 .dirty_folio = nilfs_dirty_folio,
304 .readahead = nilfs_readahead,
305 .write_begin = nilfs_write_begin,
306 .write_end = nilfs_write_end,
307 .invalidate_folio = block_invalidate_folio,
308 .direct_IO = nilfs_direct_IO,
309 .is_partially_uptodate = block_is_partially_uptodate,
312 static int nilfs_insert_inode_locked(struct inode *inode,
313 struct nilfs_root *root,
316 struct nilfs_iget_args args = {
317 .ino = ino, .root = root, .cno = 0, .for_gc = false,
318 .for_btnc = false, .for_shadow = false
321 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
324 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
326 struct super_block *sb = dir->i_sb;
327 struct the_nilfs *nilfs = sb->s_fs_info;
329 struct nilfs_inode_info *ii;
330 struct nilfs_root *root;
334 inode = new_inode(sb);
335 if (unlikely(!inode))
338 mapping_set_gfp_mask(inode->i_mapping,
339 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
341 root = NILFS_I(dir)->i_root;
343 ii->i_state = BIT(NILFS_I_NEW);
346 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
348 goto failed_ifile_create_inode;
349 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
351 atomic64_inc(&root->inodes_count);
352 inode_init_owner(&init_user_ns, inode, dir, mode);
354 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
356 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
357 err = nilfs_bmap_read(ii->i_bmap, NULL);
359 goto failed_after_creation;
361 set_bit(NILFS_I_BMAP, &ii->i_state);
362 /* No lock is needed; iget() ensures it. */
365 ii->i_flags = nilfs_mask_flags(
366 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
368 /* ii->i_file_acl = 0; */
369 /* ii->i_dir_acl = 0; */
370 ii->i_dir_start_lookup = 0;
371 nilfs_set_inode_flags(inode);
372 spin_lock(&nilfs->ns_next_gen_lock);
373 inode->i_generation = nilfs->ns_next_generation++;
374 spin_unlock(&nilfs->ns_next_gen_lock);
375 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
377 goto failed_after_creation;
380 err = nilfs_init_acl(inode, dir);
383 * Never occur. When supporting nilfs_init_acl(),
384 * proper cancellation of above jobs should be considered.
386 goto failed_after_creation;
390 failed_after_creation:
392 if (inode->i_state & I_NEW)
393 unlock_new_inode(inode);
395 * raw_inode will be deleted through
396 * nilfs_evict_inode().
400 failed_ifile_create_inode:
401 make_bad_inode(inode);
407 void nilfs_set_inode_flags(struct inode *inode)
409 unsigned int flags = NILFS_I(inode)->i_flags;
410 unsigned int new_fl = 0;
412 if (flags & FS_SYNC_FL)
414 if (flags & FS_APPEND_FL)
416 if (flags & FS_IMMUTABLE_FL)
417 new_fl |= S_IMMUTABLE;
418 if (flags & FS_NOATIME_FL)
420 if (flags & FS_DIRSYNC_FL)
422 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
423 S_NOATIME | S_DIRSYNC);
426 int nilfs_read_inode_common(struct inode *inode,
427 struct nilfs_inode *raw_inode)
429 struct nilfs_inode_info *ii = NILFS_I(inode);
432 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
433 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
434 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
435 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
436 inode->i_size = le64_to_cpu(raw_inode->i_size);
437 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
438 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
439 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
440 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
441 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
442 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
443 if (inode->i_nlink == 0)
444 return -ESTALE; /* this inode is deleted */
446 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
447 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
449 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
450 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
451 0 : le32_to_cpu(raw_inode->i_dir_acl);
453 ii->i_dir_start_lookup = 0;
454 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
456 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
457 S_ISLNK(inode->i_mode)) {
458 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
461 set_bit(NILFS_I_BMAP, &ii->i_state);
462 /* No lock is needed; iget() ensures it. */
467 static int __nilfs_read_inode(struct super_block *sb,
468 struct nilfs_root *root, unsigned long ino,
471 struct the_nilfs *nilfs = sb->s_fs_info;
472 struct buffer_head *bh;
473 struct nilfs_inode *raw_inode;
476 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
477 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
481 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
483 err = nilfs_read_inode_common(inode, raw_inode);
487 if (S_ISREG(inode->i_mode)) {
488 inode->i_op = &nilfs_file_inode_operations;
489 inode->i_fop = &nilfs_file_operations;
490 inode->i_mapping->a_ops = &nilfs_aops;
491 } else if (S_ISDIR(inode->i_mode)) {
492 inode->i_op = &nilfs_dir_inode_operations;
493 inode->i_fop = &nilfs_dir_operations;
494 inode->i_mapping->a_ops = &nilfs_aops;
495 } else if (S_ISLNK(inode->i_mode)) {
496 inode->i_op = &nilfs_symlink_inode_operations;
497 inode_nohighmem(inode);
498 inode->i_mapping->a_ops = &nilfs_aops;
500 inode->i_op = &nilfs_special_inode_operations;
502 inode, inode->i_mode,
503 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
505 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
507 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
508 nilfs_set_inode_flags(inode);
509 mapping_set_gfp_mask(inode->i_mapping,
510 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
514 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
518 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
522 static int nilfs_iget_test(struct inode *inode, void *opaque)
524 struct nilfs_iget_args *args = opaque;
525 struct nilfs_inode_info *ii;
527 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
531 if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
534 } else if (args->for_btnc) {
537 if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
538 if (!args->for_shadow)
540 } else if (args->for_shadow) {
544 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
545 return !args->for_gc;
547 return args->for_gc && args->cno == ii->i_cno;
550 static int nilfs_iget_set(struct inode *inode, void *opaque)
552 struct nilfs_iget_args *args = opaque;
554 inode->i_ino = args->ino;
555 NILFS_I(inode)->i_cno = args->cno;
556 NILFS_I(inode)->i_root = args->root;
557 if (args->root && args->ino == NILFS_ROOT_INO)
558 nilfs_get_root(args->root);
561 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
563 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
564 if (args->for_shadow)
565 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
569 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
572 struct nilfs_iget_args args = {
573 .ino = ino, .root = root, .cno = 0, .for_gc = false,
574 .for_btnc = false, .for_shadow = false
577 return ilookup5(sb, ino, nilfs_iget_test, &args);
580 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
583 struct nilfs_iget_args args = {
584 .ino = ino, .root = root, .cno = 0, .for_gc = false,
585 .for_btnc = false, .for_shadow = false
588 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
591 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
597 inode = nilfs_iget_locked(sb, root, ino);
598 if (unlikely(!inode))
599 return ERR_PTR(-ENOMEM);
600 if (!(inode->i_state & I_NEW))
603 err = __nilfs_read_inode(sb, root, ino, inode);
608 unlock_new_inode(inode);
612 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
615 struct nilfs_iget_args args = {
616 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
617 .for_btnc = false, .for_shadow = false
622 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
623 if (unlikely(!inode))
624 return ERR_PTR(-ENOMEM);
625 if (!(inode->i_state & I_NEW))
628 err = nilfs_init_gcinode(inode);
633 unlock_new_inode(inode);
638 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
639 * @inode: inode object
641 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
642 * or does nothing if the inode already has it. This function allocates
643 * an additional inode to maintain page cache of B-tree nodes one-on-one.
645 * Return Value: On success, 0 is returned. On errors, one of the following
646 * negative error code is returned.
648 * %-ENOMEM - Insufficient memory available.
650 int nilfs_attach_btree_node_cache(struct inode *inode)
652 struct nilfs_inode_info *ii = NILFS_I(inode);
653 struct inode *btnc_inode;
654 struct nilfs_iget_args args;
656 if (ii->i_assoc_inode)
659 args.ino = inode->i_ino;
660 args.root = ii->i_root;
661 args.cno = ii->i_cno;
662 args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
663 args.for_btnc = true;
664 args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
666 btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
667 nilfs_iget_set, &args);
668 if (unlikely(!btnc_inode))
670 if (btnc_inode->i_state & I_NEW) {
671 nilfs_init_btnc_inode(btnc_inode);
672 unlock_new_inode(btnc_inode);
674 NILFS_I(btnc_inode)->i_assoc_inode = inode;
675 NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
676 ii->i_assoc_inode = btnc_inode;
682 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
683 * @inode: inode object
685 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
686 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
688 void nilfs_detach_btree_node_cache(struct inode *inode)
690 struct nilfs_inode_info *ii = NILFS_I(inode);
691 struct inode *btnc_inode = ii->i_assoc_inode;
694 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
695 ii->i_assoc_inode = NULL;
701 * nilfs_iget_for_shadow - obtain inode for shadow mapping
702 * @inode: inode object that uses shadow mapping
704 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
705 * caches for shadow mapping. The page cache for data pages is set up
706 * in one inode and the one for b-tree node pages is set up in the
707 * other inode, which is attached to the former inode.
709 * Return Value: On success, a pointer to the inode for data pages is
710 * returned. On errors, one of the following negative error code is returned
713 * %-ENOMEM - Insufficient memory available.
715 struct inode *nilfs_iget_for_shadow(struct inode *inode)
717 struct nilfs_iget_args args = {
718 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
719 .for_btnc = false, .for_shadow = true
721 struct inode *s_inode;
724 s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
725 nilfs_iget_set, &args);
726 if (unlikely(!s_inode))
727 return ERR_PTR(-ENOMEM);
728 if (!(s_inode->i_state & I_NEW))
731 NILFS_I(s_inode)->i_flags = 0;
732 memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
733 mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
735 err = nilfs_attach_btree_node_cache(s_inode);
737 iget_failed(s_inode);
740 unlock_new_inode(s_inode);
744 void nilfs_write_inode_common(struct inode *inode,
745 struct nilfs_inode *raw_inode, int has_bmap)
747 struct nilfs_inode_info *ii = NILFS_I(inode);
749 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
750 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
751 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
752 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
753 raw_inode->i_size = cpu_to_le64(inode->i_size);
754 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
755 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
756 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
757 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
758 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
760 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
761 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
763 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
764 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
766 /* zero-fill unused portion in the case of super root block */
767 raw_inode->i_xattr = 0;
768 raw_inode->i_pad = 0;
769 memset((void *)raw_inode + sizeof(*raw_inode), 0,
770 nilfs->ns_inode_size - sizeof(*raw_inode));
774 nilfs_bmap_write(ii->i_bmap, raw_inode);
775 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
776 raw_inode->i_device_code =
777 cpu_to_le64(huge_encode_dev(inode->i_rdev));
779 * When extending inode, nilfs->ns_inode_size should be checked
780 * for substitutions of appended fields.
784 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
786 ino_t ino = inode->i_ino;
787 struct nilfs_inode_info *ii = NILFS_I(inode);
788 struct inode *ifile = ii->i_root->ifile;
789 struct nilfs_inode *raw_inode;
791 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
793 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
794 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
795 if (flags & I_DIRTY_DATASYNC)
796 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
798 nilfs_write_inode_common(inode, raw_inode, 0);
800 * XXX: call with has_bmap = 0 is a workaround to avoid
801 * deadlock of bmap. This delays update of i_bmap to just
805 nilfs_ifile_unmap_inode(ifile, ino, ibh);
808 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
810 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
816 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
819 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
828 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
829 ret = nilfs_bmap_truncate(ii->i_bmap, b);
830 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
831 if (!ret || (ret == -ENOMEM &&
832 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
836 nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
837 ret, ii->vfs_inode.i_ino);
840 void nilfs_truncate(struct inode *inode)
842 unsigned long blkoff;
843 unsigned int blocksize;
844 struct nilfs_transaction_info ti;
845 struct super_block *sb = inode->i_sb;
846 struct nilfs_inode_info *ii = NILFS_I(inode);
848 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
850 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
853 blocksize = sb->s_blocksize;
854 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
855 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
857 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
859 nilfs_truncate_bmap(ii, blkoff);
861 inode->i_mtime = inode->i_ctime = current_time(inode);
863 nilfs_set_transaction_flag(NILFS_TI_SYNC);
865 nilfs_mark_inode_dirty(inode);
866 nilfs_set_file_dirty(inode, 0);
867 nilfs_transaction_commit(sb);
869 * May construct a logical segment and may fail in sync mode.
870 * But truncate has no return value.
874 static void nilfs_clear_inode(struct inode *inode)
876 struct nilfs_inode_info *ii = NILFS_I(inode);
879 * Free resources allocated in nilfs_read_inode(), here.
881 BUG_ON(!list_empty(&ii->i_dirty));
885 if (nilfs_is_metadata_file_inode(inode))
886 nilfs_mdt_clear(inode);
888 if (test_bit(NILFS_I_BMAP, &ii->i_state))
889 nilfs_bmap_clear(ii->i_bmap);
891 if (!test_bit(NILFS_I_BTNC, &ii->i_state))
892 nilfs_detach_btree_node_cache(inode);
894 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
895 nilfs_put_root(ii->i_root);
898 void nilfs_evict_inode(struct inode *inode)
900 struct nilfs_transaction_info ti;
901 struct super_block *sb = inode->i_sb;
902 struct nilfs_inode_info *ii = NILFS_I(inode);
905 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
906 truncate_inode_pages_final(&inode->i_data);
908 nilfs_clear_inode(inode);
911 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
913 truncate_inode_pages_final(&inode->i_data);
915 /* TODO: some of the following operations may fail. */
916 nilfs_truncate_bmap(ii, 0);
917 nilfs_mark_inode_dirty(inode);
920 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
922 atomic64_dec(&ii->i_root->inodes_count);
924 nilfs_clear_inode(inode);
927 nilfs_set_transaction_flag(NILFS_TI_SYNC);
928 nilfs_transaction_commit(sb);
930 * May construct a logical segment and may fail in sync mode.
931 * But delete_inode has no return value.
935 int nilfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
938 struct nilfs_transaction_info ti;
939 struct inode *inode = d_inode(dentry);
940 struct super_block *sb = inode->i_sb;
943 err = setattr_prepare(&init_user_ns, dentry, iattr);
947 err = nilfs_transaction_begin(sb, &ti, 0);
951 if ((iattr->ia_valid & ATTR_SIZE) &&
952 iattr->ia_size != i_size_read(inode)) {
953 inode_dio_wait(inode);
954 truncate_setsize(inode, iattr->ia_size);
955 nilfs_truncate(inode);
958 setattr_copy(&init_user_ns, inode, iattr);
959 mark_inode_dirty(inode);
961 if (iattr->ia_valid & ATTR_MODE) {
962 err = nilfs_acl_chmod(inode);
967 return nilfs_transaction_commit(sb);
970 nilfs_transaction_abort(sb);
974 int nilfs_permission(struct user_namespace *mnt_userns, struct inode *inode,
977 struct nilfs_root *root = NILFS_I(inode)->i_root;
979 if ((mask & MAY_WRITE) && root &&
980 root->cno != NILFS_CPTREE_CURRENT_CNO)
981 return -EROFS; /* snapshot is not writable */
983 return generic_permission(&init_user_ns, inode, mask);
986 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
988 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
989 struct nilfs_inode_info *ii = NILFS_I(inode);
992 spin_lock(&nilfs->ns_inode_lock);
993 if (ii->i_bh == NULL) {
994 spin_unlock(&nilfs->ns_inode_lock);
995 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
999 spin_lock(&nilfs->ns_inode_lock);
1000 if (ii->i_bh == NULL)
1010 spin_unlock(&nilfs->ns_inode_lock);
1014 int nilfs_inode_dirty(struct inode *inode)
1016 struct nilfs_inode_info *ii = NILFS_I(inode);
1017 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1020 if (!list_empty(&ii->i_dirty)) {
1021 spin_lock(&nilfs->ns_inode_lock);
1022 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1023 test_bit(NILFS_I_BUSY, &ii->i_state);
1024 spin_unlock(&nilfs->ns_inode_lock);
1029 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1031 struct nilfs_inode_info *ii = NILFS_I(inode);
1032 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1034 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1036 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1039 spin_lock(&nilfs->ns_inode_lock);
1040 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1041 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1043 * Because this routine may race with nilfs_dispose_list(),
1044 * we have to check NILFS_I_QUEUED here, too.
1046 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1048 * This will happen when somebody is freeing
1051 nilfs_warn(inode->i_sb,
1052 "cannot set file dirty (ino=%lu): the file is being freed",
1054 spin_unlock(&nilfs->ns_inode_lock);
1056 * NILFS_I_DIRTY may remain for
1060 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1061 set_bit(NILFS_I_QUEUED, &ii->i_state);
1063 spin_unlock(&nilfs->ns_inode_lock);
1067 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1069 struct buffer_head *ibh;
1072 err = nilfs_load_inode_block(inode, &ibh);
1073 if (unlikely(err)) {
1074 nilfs_warn(inode->i_sb,
1075 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1079 nilfs_update_inode(inode, ibh, flags);
1080 mark_buffer_dirty(ibh);
1081 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1087 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1088 * @inode: inode of the file to be registered.
1089 * @flags: flags to determine the dirty state of the inode
1091 * nilfs_dirty_inode() loads a inode block containing the specified
1092 * @inode and copies data from a nilfs_inode to a corresponding inode
1093 * entry in the inode block. This operation is excluded from the segment
1094 * construction. This function can be called both as a single operation
1095 * and as a part of indivisible file operations.
1097 void nilfs_dirty_inode(struct inode *inode, int flags)
1099 struct nilfs_transaction_info ti;
1100 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
1102 if (is_bad_inode(inode)) {
1103 nilfs_warn(inode->i_sb,
1104 "tried to mark bad_inode dirty. ignored.");
1109 nilfs_mdt_mark_dirty(inode);
1112 nilfs_transaction_begin(inode->i_sb, &ti, 0);
1113 __nilfs_mark_inode_dirty(inode, flags);
1114 nilfs_transaction_commit(inode->i_sb); /* never fails */
1117 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1118 __u64 start, __u64 len)
1120 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1121 __u64 logical = 0, phys = 0, size = 0;
1124 sector_t blkoff, end_blkoff;
1125 sector_t delalloc_blkoff;
1126 unsigned long delalloc_blklen;
1127 unsigned int blkbits = inode->i_blkbits;
1130 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
1136 isize = i_size_read(inode);
1138 blkoff = start >> blkbits;
1139 end_blkoff = (start + len - 1) >> blkbits;
1141 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1146 unsigned int maxblocks;
1148 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1150 /* End of the current extent */
1151 ret = fiemap_fill_next_extent(
1152 fieinfo, logical, phys, size, flags);
1156 if (blkoff > end_blkoff)
1159 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1160 logical = blkoff << blkbits;
1162 size = delalloc_blklen << blkbits;
1164 blkoff = delalloc_blkoff + delalloc_blklen;
1165 delalloc_blklen = nilfs_find_uncommitted_extent(
1166 inode, blkoff, &delalloc_blkoff);
1171 * Limit the number of blocks that we look up so as
1172 * not to get into the next delayed allocation extent.
1174 maxblocks = INT_MAX;
1175 if (delalloc_blklen)
1176 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1180 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1181 n = nilfs_bmap_lookup_contig(
1182 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1183 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1188 if (unlikely(n != -ENOENT))
1193 past_eof = ((blkoff << blkbits) >= isize);
1196 /* End of the current extent */
1199 flags |= FIEMAP_EXTENT_LAST;
1201 ret = fiemap_fill_next_extent(
1202 fieinfo, logical, phys, size, flags);
1207 if (blkoff > end_blkoff || past_eof)
1211 if (phys && blkphy << blkbits == phys + size) {
1212 /* The current extent goes on */
1213 size += n << blkbits;
1215 /* Terminate the current extent */
1216 ret = fiemap_fill_next_extent(
1217 fieinfo, logical, phys, size,
1219 if (ret || blkoff > end_blkoff)
1222 /* Start another extent */
1223 flags = FIEMAP_EXTENT_MERGED;
1224 logical = blkoff << blkbits;
1225 phys = blkphy << blkbits;
1226 size = n << blkbits;
1229 /* Start a new extent */
1230 flags = FIEMAP_EXTENT_MERGED;
1231 logical = blkoff << blkbits;
1232 phys = blkphy << blkbits;
1233 size = n << blkbits;
1240 /* If ret is 1 then we just hit the end of the extent array */
1244 inode_unlock(inode);