1 // SPDX-License-Identifier: GPL-2.0+
3 * inode.c - 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_readpage() - implement readpage() method of nilfs_aops {}
144 * address_space_operations.
145 * @file - file struct of the file to be read
146 * @page - the page to be read
148 static int nilfs_readpage(struct file *file, struct page *page)
150 return mpage_readpage(page, 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 int nilfs_set_page_dirty(struct page *page)
208 struct inode *inode = page->mapping->host;
209 int ret = __set_page_dirty_nobuffers(page);
211 if (page_has_buffers(page)) {
212 unsigned int nr_dirty = 0;
213 struct buffer_head *bh, *head;
216 * This page is locked by callers, and no other thread
217 * concurrently marks its buffers dirty since they are
218 * only dirtied through routines in fs/buffer.c in
219 * which call sites of mark_buffer_dirty are protected
222 bh = head = page_buffers(page);
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);
233 nilfs_set_file_dirty(inode, nr_dirty);
235 unsigned int nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
237 nilfs_set_file_dirty(inode, nr_dirty);
242 void nilfs_write_failed(struct address_space *mapping, loff_t to)
244 struct inode *inode = mapping->host;
246 if (to > inode->i_size) {
247 truncate_pagecache(inode, inode->i_size);
248 nilfs_truncate(inode);
252 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
253 loff_t pos, unsigned len, unsigned flags,
254 struct page **pagep, void **fsdata)
257 struct inode *inode = mapping->host;
258 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
263 err = block_write_begin(mapping, pos, len, flags, pagep,
266 nilfs_write_failed(mapping, pos + len);
267 nilfs_transaction_abort(inode->i_sb);
272 static int nilfs_write_end(struct file *file, struct address_space *mapping,
273 loff_t pos, unsigned len, unsigned copied,
274 struct page *page, void *fsdata)
276 struct inode *inode = mapping->host;
277 unsigned int start = pos & (PAGE_SIZE - 1);
278 unsigned int nr_dirty;
281 nr_dirty = nilfs_page_count_clean_buffers(page, start,
283 copied = generic_write_end(file, mapping, pos, len, copied, page,
285 nilfs_set_file_dirty(inode, nr_dirty);
286 err = nilfs_transaction_commit(inode->i_sb);
287 return err ? : copied;
291 nilfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
293 struct inode *inode = file_inode(iocb->ki_filp);
295 if (iov_iter_rw(iter) == WRITE)
298 /* Needs synchronization with the cleaner */
299 return blockdev_direct_IO(iocb, inode, iter, nilfs_get_block);
302 const struct address_space_operations nilfs_aops = {
303 .writepage = nilfs_writepage,
304 .readpage = nilfs_readpage,
305 .writepages = nilfs_writepages,
306 .set_page_dirty = nilfs_set_page_dirty,
307 .readahead = nilfs_readahead,
308 .write_begin = nilfs_write_begin,
309 .write_end = nilfs_write_end,
310 /* .releasepage = nilfs_releasepage, */
311 .invalidatepage = block_invalidatepage,
312 .direct_IO = nilfs_direct_IO,
313 .is_partially_uptodate = block_is_partially_uptodate,
316 static int nilfs_insert_inode_locked(struct inode *inode,
317 struct nilfs_root *root,
320 struct nilfs_iget_args args = {
321 .ino = ino, .root = root, .cno = 0, .for_gc = false,
322 .for_btnc = false, .for_shadow = false
325 return insert_inode_locked4(inode, ino, nilfs_iget_test, &args);
328 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
330 struct super_block *sb = dir->i_sb;
331 struct the_nilfs *nilfs = sb->s_fs_info;
333 struct nilfs_inode_info *ii;
334 struct nilfs_root *root;
335 struct buffer_head *bh;
339 inode = new_inode(sb);
340 if (unlikely(!inode))
343 mapping_set_gfp_mask(inode->i_mapping,
344 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
346 root = NILFS_I(dir)->i_root;
348 ii->i_state = BIT(NILFS_I_NEW);
351 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
353 goto failed_ifile_create_inode;
354 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
356 if (unlikely(ino < NILFS_USER_INO)) {
358 "inode bitmap is inconsistent for reserved inodes");
361 err = nilfs_ifile_create_inode(root->ifile, &ino, &bh);
363 goto failed_ifile_create_inode;
364 } while (ino < NILFS_USER_INO);
366 nilfs_info(sb, "repaired inode bitmap for reserved inodes");
370 atomic64_inc(&root->inodes_count);
371 inode_init_owner(inode, dir, mode);
373 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
375 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
376 err = nilfs_bmap_read(ii->i_bmap, NULL);
378 goto failed_after_creation;
380 set_bit(NILFS_I_BMAP, &ii->i_state);
381 /* No lock is needed; iget() ensures it. */
384 ii->i_flags = nilfs_mask_flags(
385 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
387 /* ii->i_file_acl = 0; */
388 /* ii->i_dir_acl = 0; */
389 ii->i_dir_start_lookup = 0;
390 nilfs_set_inode_flags(inode);
391 spin_lock(&nilfs->ns_next_gen_lock);
392 inode->i_generation = nilfs->ns_next_generation++;
393 spin_unlock(&nilfs->ns_next_gen_lock);
394 if (nilfs_insert_inode_locked(inode, root, ino) < 0) {
396 goto failed_after_creation;
399 err = nilfs_init_acl(inode, dir);
402 * Never occur. When supporting nilfs_init_acl(),
403 * proper cancellation of above jobs should be considered.
405 goto failed_after_creation;
409 failed_after_creation:
411 if (inode->i_state & I_NEW)
412 unlock_new_inode(inode);
414 * raw_inode will be deleted through
415 * nilfs_evict_inode().
419 failed_ifile_create_inode:
420 make_bad_inode(inode);
426 void nilfs_set_inode_flags(struct inode *inode)
428 unsigned int flags = NILFS_I(inode)->i_flags;
429 unsigned int new_fl = 0;
431 if (flags & FS_SYNC_FL)
433 if (flags & FS_APPEND_FL)
435 if (flags & FS_IMMUTABLE_FL)
436 new_fl |= S_IMMUTABLE;
437 if (flags & FS_NOATIME_FL)
439 if (flags & FS_DIRSYNC_FL)
441 inode_set_flags(inode, new_fl, S_SYNC | S_APPEND | S_IMMUTABLE |
442 S_NOATIME | S_DIRSYNC);
445 int nilfs_read_inode_common(struct inode *inode,
446 struct nilfs_inode *raw_inode)
448 struct nilfs_inode_info *ii = NILFS_I(inode);
451 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
452 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
453 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
454 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
455 inode->i_size = le64_to_cpu(raw_inode->i_size);
456 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
457 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
458 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
459 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
460 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
461 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
462 if (nilfs_is_metadata_file_inode(inode) && !S_ISREG(inode->i_mode))
463 return -EIO; /* this inode is for metadata and corrupted */
464 if (inode->i_nlink == 0)
465 return -ESTALE; /* this inode is deleted */
467 inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
468 ii->i_flags = le32_to_cpu(raw_inode->i_flags);
470 ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
471 ii->i_dir_acl = S_ISREG(inode->i_mode) ?
472 0 : le32_to_cpu(raw_inode->i_dir_acl);
474 ii->i_dir_start_lookup = 0;
475 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
477 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
478 S_ISLNK(inode->i_mode)) {
479 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
482 set_bit(NILFS_I_BMAP, &ii->i_state);
483 /* No lock is needed; iget() ensures it. */
488 static int __nilfs_read_inode(struct super_block *sb,
489 struct nilfs_root *root, unsigned long ino,
492 struct the_nilfs *nilfs = sb->s_fs_info;
493 struct buffer_head *bh;
494 struct nilfs_inode *raw_inode;
497 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
498 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
502 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
504 err = nilfs_read_inode_common(inode, raw_inode);
508 if (S_ISREG(inode->i_mode)) {
509 inode->i_op = &nilfs_file_inode_operations;
510 inode->i_fop = &nilfs_file_operations;
511 inode->i_mapping->a_ops = &nilfs_aops;
512 } else if (S_ISDIR(inode->i_mode)) {
513 inode->i_op = &nilfs_dir_inode_operations;
514 inode->i_fop = &nilfs_dir_operations;
515 inode->i_mapping->a_ops = &nilfs_aops;
516 } else if (S_ISLNK(inode->i_mode)) {
517 inode->i_op = &nilfs_symlink_inode_operations;
518 inode_nohighmem(inode);
519 inode->i_mapping->a_ops = &nilfs_aops;
521 inode->i_op = &nilfs_special_inode_operations;
523 inode, inode->i_mode,
524 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
526 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
528 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
529 nilfs_set_inode_flags(inode);
530 mapping_set_gfp_mask(inode->i_mapping,
531 mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
535 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
539 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
543 static int nilfs_iget_test(struct inode *inode, void *opaque)
545 struct nilfs_iget_args *args = opaque;
546 struct nilfs_inode_info *ii;
548 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
552 if (test_bit(NILFS_I_BTNC, &ii->i_state)) {
555 } else if (args->for_btnc) {
558 if (test_bit(NILFS_I_SHADOW, &ii->i_state)) {
559 if (!args->for_shadow)
561 } else if (args->for_shadow) {
565 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
566 return !args->for_gc;
568 return args->for_gc && args->cno == ii->i_cno;
571 static int nilfs_iget_set(struct inode *inode, void *opaque)
573 struct nilfs_iget_args *args = opaque;
575 inode->i_ino = args->ino;
576 NILFS_I(inode)->i_cno = args->cno;
577 NILFS_I(inode)->i_root = args->root;
578 if (args->root && args->ino == NILFS_ROOT_INO)
579 nilfs_get_root(args->root);
582 NILFS_I(inode)->i_state = BIT(NILFS_I_GCINODE);
584 NILFS_I(inode)->i_state |= BIT(NILFS_I_BTNC);
585 if (args->for_shadow)
586 NILFS_I(inode)->i_state |= BIT(NILFS_I_SHADOW);
590 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
593 struct nilfs_iget_args args = {
594 .ino = ino, .root = root, .cno = 0, .for_gc = false,
595 .for_btnc = false, .for_shadow = false
598 return ilookup5(sb, ino, nilfs_iget_test, &args);
601 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
604 struct nilfs_iget_args args = {
605 .ino = ino, .root = root, .cno = 0, .for_gc = false,
606 .for_btnc = false, .for_shadow = false
609 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
612 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
618 inode = nilfs_iget_locked(sb, root, ino);
619 if (unlikely(!inode))
620 return ERR_PTR(-ENOMEM);
621 if (!(inode->i_state & I_NEW))
624 err = __nilfs_read_inode(sb, root, ino, inode);
629 unlock_new_inode(inode);
633 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
636 struct nilfs_iget_args args = {
637 .ino = ino, .root = NULL, .cno = cno, .for_gc = true,
638 .for_btnc = false, .for_shadow = false
643 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
644 if (unlikely(!inode))
645 return ERR_PTR(-ENOMEM);
646 if (!(inode->i_state & I_NEW))
649 err = nilfs_init_gcinode(inode);
654 unlock_new_inode(inode);
659 * nilfs_attach_btree_node_cache - attach a B-tree node cache to the inode
660 * @inode: inode object
662 * nilfs_attach_btree_node_cache() attaches a B-tree node cache to @inode,
663 * or does nothing if the inode already has it. This function allocates
664 * an additional inode to maintain page cache of B-tree nodes one-on-one.
666 * Return Value: On success, 0 is returned. On errors, one of the following
667 * negative error code is returned.
669 * %-ENOMEM - Insufficient memory available.
671 int nilfs_attach_btree_node_cache(struct inode *inode)
673 struct nilfs_inode_info *ii = NILFS_I(inode);
674 struct inode *btnc_inode;
675 struct nilfs_iget_args args;
677 if (ii->i_assoc_inode)
680 args.ino = inode->i_ino;
681 args.root = ii->i_root;
682 args.cno = ii->i_cno;
683 args.for_gc = test_bit(NILFS_I_GCINODE, &ii->i_state) != 0;
684 args.for_btnc = true;
685 args.for_shadow = test_bit(NILFS_I_SHADOW, &ii->i_state) != 0;
687 btnc_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
688 nilfs_iget_set, &args);
689 if (unlikely(!btnc_inode))
691 if (btnc_inode->i_state & I_NEW) {
692 nilfs_init_btnc_inode(btnc_inode);
693 unlock_new_inode(btnc_inode);
695 NILFS_I(btnc_inode)->i_assoc_inode = inode;
696 NILFS_I(btnc_inode)->i_bmap = ii->i_bmap;
697 ii->i_assoc_inode = btnc_inode;
703 * nilfs_detach_btree_node_cache - detach the B-tree node cache from the inode
704 * @inode: inode object
706 * nilfs_detach_btree_node_cache() detaches the B-tree node cache and its
707 * holder inode bound to @inode, or does nothing if @inode doesn't have it.
709 void nilfs_detach_btree_node_cache(struct inode *inode)
711 struct nilfs_inode_info *ii = NILFS_I(inode);
712 struct inode *btnc_inode = ii->i_assoc_inode;
715 NILFS_I(btnc_inode)->i_assoc_inode = NULL;
716 ii->i_assoc_inode = NULL;
722 * nilfs_iget_for_shadow - obtain inode for shadow mapping
723 * @inode: inode object that uses shadow mapping
725 * nilfs_iget_for_shadow() allocates a pair of inodes that holds page
726 * caches for shadow mapping. The page cache for data pages is set up
727 * in one inode and the one for b-tree node pages is set up in the
728 * other inode, which is attached to the former inode.
730 * Return Value: On success, a pointer to the inode for data pages is
731 * returned. On errors, one of the following negative error code is returned
734 * %-ENOMEM - Insufficient memory available.
736 struct inode *nilfs_iget_for_shadow(struct inode *inode)
738 struct nilfs_iget_args args = {
739 .ino = inode->i_ino, .root = NULL, .cno = 0, .for_gc = false,
740 .for_btnc = false, .for_shadow = true
742 struct inode *s_inode;
745 s_inode = iget5_locked(inode->i_sb, inode->i_ino, nilfs_iget_test,
746 nilfs_iget_set, &args);
747 if (unlikely(!s_inode))
748 return ERR_PTR(-ENOMEM);
749 if (!(s_inode->i_state & I_NEW))
752 NILFS_I(s_inode)->i_flags = 0;
753 memset(NILFS_I(s_inode)->i_bmap, 0, sizeof(struct nilfs_bmap));
754 mapping_set_gfp_mask(s_inode->i_mapping, GFP_NOFS);
756 err = nilfs_attach_btree_node_cache(s_inode);
758 iget_failed(s_inode);
761 unlock_new_inode(s_inode);
765 void nilfs_write_inode_common(struct inode *inode,
766 struct nilfs_inode *raw_inode, int has_bmap)
768 struct nilfs_inode_info *ii = NILFS_I(inode);
770 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
771 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
772 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
773 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
774 raw_inode->i_size = cpu_to_le64(inode->i_size);
775 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
776 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
777 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
778 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
779 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
781 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
782 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
784 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
785 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
787 /* zero-fill unused portion in the case of super root block */
788 raw_inode->i_xattr = 0;
789 raw_inode->i_pad = 0;
790 memset((void *)raw_inode + sizeof(*raw_inode), 0,
791 nilfs->ns_inode_size - sizeof(*raw_inode));
795 nilfs_bmap_write(ii->i_bmap, raw_inode);
796 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
797 raw_inode->i_device_code =
798 cpu_to_le64(huge_encode_dev(inode->i_rdev));
800 * When extending inode, nilfs->ns_inode_size should be checked
801 * for substitutions of appended fields.
805 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
807 ino_t ino = inode->i_ino;
808 struct nilfs_inode_info *ii = NILFS_I(inode);
809 struct inode *ifile = ii->i_root->ifile;
810 struct nilfs_inode *raw_inode;
812 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
814 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
815 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
816 if (flags & I_DIRTY_DATASYNC)
817 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
819 nilfs_write_inode_common(inode, raw_inode, 0);
821 * XXX: call with has_bmap = 0 is a workaround to avoid
822 * deadlock of bmap. This delays update of i_bmap to just
826 nilfs_ifile_unmap_inode(ifile, ino, ibh);
829 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
831 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
837 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
840 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
849 b -= min_t(__u64, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
850 ret = nilfs_bmap_truncate(ii->i_bmap, b);
851 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
852 if (!ret || (ret == -ENOMEM &&
853 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
857 nilfs_warn(ii->vfs_inode.i_sb, "error %d truncating bmap (ino=%lu)",
858 ret, ii->vfs_inode.i_ino);
861 void nilfs_truncate(struct inode *inode)
863 unsigned long blkoff;
864 unsigned int blocksize;
865 struct nilfs_transaction_info ti;
866 struct super_block *sb = inode->i_sb;
867 struct nilfs_inode_info *ii = NILFS_I(inode);
869 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
871 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
874 blocksize = sb->s_blocksize;
875 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
876 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
878 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
880 nilfs_truncate_bmap(ii, blkoff);
882 inode->i_mtime = inode->i_ctime = current_time(inode);
884 nilfs_set_transaction_flag(NILFS_TI_SYNC);
886 nilfs_mark_inode_dirty(inode);
887 nilfs_set_file_dirty(inode, 0);
888 nilfs_transaction_commit(sb);
890 * May construct a logical segment and may fail in sync mode.
891 * But truncate has no return value.
895 static void nilfs_clear_inode(struct inode *inode)
897 struct nilfs_inode_info *ii = NILFS_I(inode);
900 * Free resources allocated in nilfs_read_inode(), here.
902 BUG_ON(!list_empty(&ii->i_dirty));
906 if (nilfs_is_metadata_file_inode(inode))
907 nilfs_mdt_clear(inode);
909 if (test_bit(NILFS_I_BMAP, &ii->i_state))
910 nilfs_bmap_clear(ii->i_bmap);
912 if (!test_bit(NILFS_I_BTNC, &ii->i_state))
913 nilfs_detach_btree_node_cache(inode);
915 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
916 nilfs_put_root(ii->i_root);
919 void nilfs_evict_inode(struct inode *inode)
921 struct nilfs_transaction_info ti;
922 struct super_block *sb = inode->i_sb;
923 struct nilfs_inode_info *ii = NILFS_I(inode);
924 struct the_nilfs *nilfs;
927 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
928 truncate_inode_pages_final(&inode->i_data);
930 nilfs_clear_inode(inode);
933 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
935 truncate_inode_pages_final(&inode->i_data);
937 nilfs = sb->s_fs_info;
938 if (unlikely(sb_rdonly(sb) || !nilfs->ns_writer)) {
940 * If this inode is about to be disposed after the file system
941 * has been degraded to read-only due to file system corruption
942 * or after the writer has been detached, do not make any
943 * changes that cause writes, just clear it.
944 * Do this check after read-locking ns_segctor_sem by
945 * nilfs_transaction_begin() in order to avoid a race with
946 * the writer detach operation.
949 nilfs_clear_inode(inode);
950 nilfs_transaction_abort(sb);
954 /* TODO: some of the following operations may fail. */
955 nilfs_truncate_bmap(ii, 0);
956 nilfs_mark_inode_dirty(inode);
959 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
961 atomic64_dec(&ii->i_root->inodes_count);
963 nilfs_clear_inode(inode);
966 nilfs_set_transaction_flag(NILFS_TI_SYNC);
967 nilfs_transaction_commit(sb);
969 * May construct a logical segment and may fail in sync mode.
970 * But delete_inode has no return value.
974 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
976 struct nilfs_transaction_info ti;
977 struct inode *inode = d_inode(dentry);
978 struct super_block *sb = inode->i_sb;
981 err = setattr_prepare(dentry, iattr);
985 err = nilfs_transaction_begin(sb, &ti, 0);
989 if ((iattr->ia_valid & ATTR_SIZE) &&
990 iattr->ia_size != i_size_read(inode)) {
991 inode_dio_wait(inode);
992 truncate_setsize(inode, iattr->ia_size);
993 nilfs_truncate(inode);
996 setattr_copy(inode, iattr);
997 mark_inode_dirty(inode);
999 if (iattr->ia_valid & ATTR_MODE) {
1000 err = nilfs_acl_chmod(inode);
1005 return nilfs_transaction_commit(sb);
1008 nilfs_transaction_abort(sb);
1012 int nilfs_permission(struct inode *inode, int mask)
1014 struct nilfs_root *root = NILFS_I(inode)->i_root;
1016 if ((mask & MAY_WRITE) && root &&
1017 root->cno != NILFS_CPTREE_CURRENT_CNO)
1018 return -EROFS; /* snapshot is not writable */
1020 return generic_permission(inode, mask);
1023 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
1025 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1026 struct nilfs_inode_info *ii = NILFS_I(inode);
1029 spin_lock(&nilfs->ns_inode_lock);
1030 if (ii->i_bh == NULL || unlikely(!buffer_uptodate(ii->i_bh))) {
1031 spin_unlock(&nilfs->ns_inode_lock);
1032 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
1036 spin_lock(&nilfs->ns_inode_lock);
1037 if (ii->i_bh == NULL)
1039 else if (unlikely(!buffer_uptodate(ii->i_bh))) {
1050 spin_unlock(&nilfs->ns_inode_lock);
1054 int nilfs_inode_dirty(struct inode *inode)
1056 struct nilfs_inode_info *ii = NILFS_I(inode);
1057 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1060 if (!list_empty(&ii->i_dirty)) {
1061 spin_lock(&nilfs->ns_inode_lock);
1062 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
1063 test_bit(NILFS_I_BUSY, &ii->i_state);
1064 spin_unlock(&nilfs->ns_inode_lock);
1069 int nilfs_set_file_dirty(struct inode *inode, unsigned int nr_dirty)
1071 struct nilfs_inode_info *ii = NILFS_I(inode);
1072 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1074 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
1076 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
1079 spin_lock(&nilfs->ns_inode_lock);
1080 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
1081 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
1083 * Because this routine may race with nilfs_dispose_list(),
1084 * we have to check NILFS_I_QUEUED here, too.
1086 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
1088 * This will happen when somebody is freeing
1091 nilfs_warn(inode->i_sb,
1092 "cannot set file dirty (ino=%lu): the file is being freed",
1094 spin_unlock(&nilfs->ns_inode_lock);
1096 * NILFS_I_DIRTY may remain for
1100 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
1101 set_bit(NILFS_I_QUEUED, &ii->i_state);
1103 spin_unlock(&nilfs->ns_inode_lock);
1107 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
1109 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
1110 struct buffer_head *ibh;
1114 * Do not dirty inodes after the log writer has been detached
1115 * and its nilfs_root struct has been freed.
1117 if (unlikely(nilfs_purging(nilfs)))
1120 err = nilfs_load_inode_block(inode, &ibh);
1121 if (unlikely(err)) {
1122 nilfs_warn(inode->i_sb,
1123 "cannot mark inode dirty (ino=%lu): error %d loading inode block",
1127 nilfs_update_inode(inode, ibh, flags);
1128 mark_buffer_dirty(ibh);
1129 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
1135 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
1136 * @inode: inode of the file to be registered.
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);