2 * inode.c - NILFS inode operations.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
24 #include <linux/buffer_head.h>
25 #include <linux/gfp.h>
26 #include <linux/mpage.h>
27 #include <linux/pagemap.h>
28 #include <linux/writeback.h>
29 #include <linux/aio.h>
39 * struct nilfs_iget_args - arguments used during comparison between inodes
41 * @cno: checkpoint number
42 * @root: pointer on NILFS root object (mounted checkpoint)
43 * @for_gc: inode for GC flag
45 struct nilfs_iget_args {
48 struct nilfs_root *root;
52 void nilfs_inode_add_blocks(struct inode *inode, int n)
54 struct nilfs_root *root = NILFS_I(inode)->i_root;
56 inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
58 atomic64_add(n, &root->blocks_count);
61 void nilfs_inode_sub_blocks(struct inode *inode, int n)
63 struct nilfs_root *root = NILFS_I(inode)->i_root;
65 inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
67 atomic64_sub(n, &root->blocks_count);
71 * nilfs_get_block() - get a file block on the filesystem (callback function)
72 * @inode - inode struct of the target file
73 * @blkoff - file block number
74 * @bh_result - buffer head to be mapped on
75 * @create - indicate whether allocating the block or not when it has not
78 * This function does not issue actual read request of the specified data
79 * block. It is done by VFS.
81 int nilfs_get_block(struct inode *inode, sector_t blkoff,
82 struct buffer_head *bh_result, int create)
84 struct nilfs_inode_info *ii = NILFS_I(inode);
85 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
88 unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
90 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
91 ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
92 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
93 if (ret >= 0) { /* found */
94 map_bh(bh_result, inode->i_sb, blknum);
96 bh_result->b_size = (ret << inode->i_blkbits);
99 /* data block was not found */
100 if (ret == -ENOENT && create) {
101 struct nilfs_transaction_info ti;
103 bh_result->b_blocknr = 0;
104 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
107 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
108 (unsigned long)bh_result);
109 if (unlikely(err != 0)) {
110 if (err == -EEXIST) {
112 * The get_block() function could be called
113 * from multiple callers for an inode.
114 * However, the page having this block must
115 * be locked in this case.
118 "nilfs_get_block: a race condition "
119 "while inserting a data block. "
120 "(inode number=%lu, file block "
123 (unsigned long long)blkoff);
126 nilfs_transaction_abort(inode->i_sb);
129 nilfs_mark_inode_dirty_sync(inode);
130 nilfs_transaction_commit(inode->i_sb); /* never fails */
131 /* Error handling should be detailed */
132 set_buffer_new(bh_result);
133 set_buffer_delay(bh_result);
134 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
136 } else if (ret == -ENOENT) {
137 /* not found is not error (e.g. hole); must return without
138 the mapped state flag. */
149 * nilfs_readpage() - implement readpage() method of nilfs_aops {}
150 * address_space_operations.
151 * @file - file struct of the file to be read
152 * @page - the page to be read
154 static int nilfs_readpage(struct file *file, struct page *page)
156 return mpage_readpage(page, nilfs_get_block);
160 * nilfs_readpages() - implement readpages() method of nilfs_aops {}
161 * address_space_operations.
162 * @file - file struct of the file to be read
163 * @mapping - address_space struct used for reading multiple pages
164 * @pages - the pages to be read
165 * @nr_pages - number of pages to be read
167 static int nilfs_readpages(struct file *file, struct address_space *mapping,
168 struct list_head *pages, unsigned nr_pages)
170 return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
173 static int nilfs_writepages(struct address_space *mapping,
174 struct writeback_control *wbc)
176 struct inode *inode = mapping->host;
179 if (inode->i_sb->s_flags & MS_RDONLY) {
180 nilfs_clear_dirty_pages(mapping, false);
184 if (wbc->sync_mode == WB_SYNC_ALL)
185 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
191 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
193 struct inode *inode = page->mapping->host;
196 if (inode->i_sb->s_flags & MS_RDONLY) {
198 * It means that filesystem was remounted in read-only
199 * mode because of error or metadata corruption. But we
200 * have dirty pages that try to be flushed in background.
201 * So, here we simply discard this dirty page.
203 nilfs_clear_dirty_page(page, false);
208 redirty_page_for_writepage(wbc, page);
211 if (wbc->sync_mode == WB_SYNC_ALL) {
212 err = nilfs_construct_segment(inode->i_sb);
215 } else if (wbc->for_reclaim)
216 nilfs_flush_segment(inode->i_sb, inode->i_ino);
221 static int nilfs_set_page_dirty(struct page *page)
223 struct inode *inode = page->mapping->host;
224 int ret = __set_page_dirty_nobuffers(page);
226 if (page_has_buffers(page)) {
227 unsigned nr_dirty = 0;
228 struct buffer_head *bh, *head;
231 * This page is locked by callers, and no other thread
232 * concurrently marks its buffers dirty since they are
233 * only dirtied through routines in fs/buffer.c in
234 * which call sites of mark_buffer_dirty are protected
237 bh = head = page_buffers(page);
239 /* Do not mark hole blocks dirty */
240 if (buffer_dirty(bh) || !buffer_mapped(bh))
243 set_buffer_dirty(bh);
245 } while (bh = bh->b_this_page, bh != head);
248 nilfs_set_file_dirty(inode, nr_dirty);
250 unsigned nr_dirty = 1 << (PAGE_CACHE_SHIFT - inode->i_blkbits);
252 nilfs_set_file_dirty(inode, nr_dirty);
257 void nilfs_write_failed(struct address_space *mapping, loff_t to)
259 struct inode *inode = mapping->host;
261 if (to > inode->i_size) {
262 truncate_pagecache(inode, inode->i_size);
263 nilfs_truncate(inode);
267 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
268 loff_t pos, unsigned len, unsigned flags,
269 struct page **pagep, void **fsdata)
272 struct inode *inode = mapping->host;
273 int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
278 err = block_write_begin(mapping, pos, len, flags, pagep,
281 nilfs_write_failed(mapping, pos + len);
282 nilfs_transaction_abort(inode->i_sb);
287 static int nilfs_write_end(struct file *file, struct address_space *mapping,
288 loff_t pos, unsigned len, unsigned copied,
289 struct page *page, void *fsdata)
291 struct inode *inode = mapping->host;
292 unsigned start = pos & (PAGE_CACHE_SIZE - 1);
296 nr_dirty = nilfs_page_count_clean_buffers(page, start,
298 copied = generic_write_end(file, mapping, pos, len, copied, page,
300 nilfs_set_file_dirty(inode, nr_dirty);
301 err = nilfs_transaction_commit(inode->i_sb);
302 return err ? : copied;
306 nilfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
309 struct file *file = iocb->ki_filp;
310 struct address_space *mapping = file->f_mapping;
311 struct inode *inode = file->f_mapping->host;
312 size_t count = iov_iter_count(iter);
318 /* Needs synchronization with the cleaner */
319 size = blockdev_direct_IO(rw, iocb, inode, iter, offset,
323 * In case of error extending write may have instantiated a few
324 * blocks outside i_size. Trim these off again.
326 if (unlikely((rw & WRITE) && size < 0)) {
327 loff_t isize = i_size_read(inode);
328 loff_t end = offset + count;
331 nilfs_write_failed(mapping, end);
337 const struct address_space_operations nilfs_aops = {
338 .writepage = nilfs_writepage,
339 .readpage = nilfs_readpage,
340 .writepages = nilfs_writepages,
341 .set_page_dirty = nilfs_set_page_dirty,
342 .readpages = nilfs_readpages,
343 .write_begin = nilfs_write_begin,
344 .write_end = nilfs_write_end,
345 /* .releasepage = nilfs_releasepage, */
346 .invalidatepage = block_invalidatepage,
347 .direct_IO = nilfs_direct_IO,
348 .is_partially_uptodate = block_is_partially_uptodate,
351 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
353 struct super_block *sb = dir->i_sb;
354 struct the_nilfs *nilfs = sb->s_fs_info;
356 struct nilfs_inode_info *ii;
357 struct nilfs_root *root;
361 inode = new_inode(sb);
362 if (unlikely(!inode))
365 mapping_set_gfp_mask(inode->i_mapping,
366 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
368 root = NILFS_I(dir)->i_root;
370 ii->i_state = 1 << NILFS_I_NEW;
373 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
375 goto failed_ifile_create_inode;
376 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
378 atomic64_inc(&root->inodes_count);
379 inode_init_owner(inode, dir, mode);
381 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
383 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
384 err = nilfs_bmap_read(ii->i_bmap, NULL);
388 set_bit(NILFS_I_BMAP, &ii->i_state);
389 /* No lock is needed; iget() ensures it. */
392 ii->i_flags = nilfs_mask_flags(
393 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
395 /* ii->i_file_acl = 0; */
396 /* ii->i_dir_acl = 0; */
397 ii->i_dir_start_lookup = 0;
398 nilfs_set_inode_flags(inode);
399 spin_lock(&nilfs->ns_next_gen_lock);
400 inode->i_generation = nilfs->ns_next_generation++;
401 spin_unlock(&nilfs->ns_next_gen_lock);
402 insert_inode_hash(inode);
404 err = nilfs_init_acl(inode, dir);
406 goto failed_acl; /* never occur. When supporting
407 nilfs_init_acl(), proper cancellation of
408 above jobs should be considered */
415 iput(inode); /* raw_inode will be deleted through
416 generic_delete_inode() */
419 failed_ifile_create_inode:
420 make_bad_inode(inode);
421 iput(inode); /* if i_nlink == 1, generic_forget_inode() will be
427 void nilfs_set_inode_flags(struct inode *inode)
429 unsigned int flags = NILFS_I(inode)->i_flags;
431 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
433 if (flags & FS_SYNC_FL)
434 inode->i_flags |= S_SYNC;
435 if (flags & FS_APPEND_FL)
436 inode->i_flags |= S_APPEND;
437 if (flags & FS_IMMUTABLE_FL)
438 inode->i_flags |= S_IMMUTABLE;
439 if (flags & FS_NOATIME_FL)
440 inode->i_flags |= S_NOATIME;
441 if (flags & FS_DIRSYNC_FL)
442 inode->i_flags |= S_DIRSYNC;
443 mapping_set_gfp_mask(inode->i_mapping,
444 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
447 int nilfs_read_inode_common(struct inode *inode,
448 struct nilfs_inode *raw_inode)
450 struct nilfs_inode_info *ii = NILFS_I(inode);
453 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
454 i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
455 i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
456 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
457 inode->i_size = le64_to_cpu(raw_inode->i_size);
458 inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
459 inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
460 inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
461 inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
462 inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
463 inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
464 if (inode->i_nlink == 0 && inode->i_mode == 0)
465 return -EINVAL; /* 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->i_mapping->a_ops = &nilfs_aops;
520 inode->i_op = &nilfs_special_inode_operations;
522 inode, inode->i_mode,
523 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
525 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
527 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
528 nilfs_set_inode_flags(inode);
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_GCINODE, &ii->i_state))
550 return !args->for_gc;
552 return args->for_gc && args->cno == ii->i_cno;
555 static int nilfs_iget_set(struct inode *inode, void *opaque)
557 struct nilfs_iget_args *args = opaque;
559 inode->i_ino = args->ino;
561 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
562 NILFS_I(inode)->i_cno = args->cno;
563 NILFS_I(inode)->i_root = NULL;
565 if (args->root && args->ino == NILFS_ROOT_INO)
566 nilfs_get_root(args->root);
567 NILFS_I(inode)->i_root = args->root;
572 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
575 struct nilfs_iget_args args = {
576 .ino = ino, .root = root, .cno = 0, .for_gc = 0
579 return ilookup5(sb, ino, nilfs_iget_test, &args);
582 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
585 struct nilfs_iget_args args = {
586 .ino = ino, .root = root, .cno = 0, .for_gc = 0
589 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
592 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
598 inode = nilfs_iget_locked(sb, root, ino);
599 if (unlikely(!inode))
600 return ERR_PTR(-ENOMEM);
601 if (!(inode->i_state & I_NEW))
604 err = __nilfs_read_inode(sb, root, ino, inode);
609 unlock_new_inode(inode);
613 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
616 struct nilfs_iget_args args = {
617 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
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);
637 void nilfs_write_inode_common(struct inode *inode,
638 struct nilfs_inode *raw_inode, int has_bmap)
640 struct nilfs_inode_info *ii = NILFS_I(inode);
642 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
643 raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
644 raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
645 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
646 raw_inode->i_size = cpu_to_le64(inode->i_size);
647 raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
648 raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
649 raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
650 raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
651 raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
653 raw_inode->i_flags = cpu_to_le32(ii->i_flags);
654 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
656 if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
657 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
659 /* zero-fill unused portion in the case of super root block */
660 raw_inode->i_xattr = 0;
661 raw_inode->i_pad = 0;
662 memset((void *)raw_inode + sizeof(*raw_inode), 0,
663 nilfs->ns_inode_size - sizeof(*raw_inode));
667 nilfs_bmap_write(ii->i_bmap, raw_inode);
668 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
669 raw_inode->i_device_code =
670 cpu_to_le64(huge_encode_dev(inode->i_rdev));
671 /* When extending inode, nilfs->ns_inode_size should be checked
672 for substitutions of appended fields */
675 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh, int flags)
677 ino_t ino = inode->i_ino;
678 struct nilfs_inode_info *ii = NILFS_I(inode);
679 struct inode *ifile = ii->i_root->ifile;
680 struct nilfs_inode *raw_inode;
682 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
684 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
685 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
686 if (flags & I_DIRTY_DATASYNC)
687 set_bit(NILFS_I_INODE_SYNC, &ii->i_state);
689 nilfs_write_inode_common(inode, raw_inode, 0);
690 /* XXX: call with has_bmap = 0 is a workaround to avoid
691 deadlock of bmap. This delays update of i_bmap to just
693 nilfs_ifile_unmap_inode(ifile, ino, ibh);
696 #define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
698 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
704 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
707 ret = nilfs_bmap_last_key(ii->i_bmap, &b);
716 b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
717 ret = nilfs_bmap_truncate(ii->i_bmap, b);
718 nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
719 if (!ret || (ret == -ENOMEM &&
720 nilfs_bmap_truncate(ii->i_bmap, b) == 0))
724 nilfs_warning(ii->vfs_inode.i_sb, __func__,
725 "failed to truncate bmap (ino=%lu, err=%d)",
726 ii->vfs_inode.i_ino, ret);
729 void nilfs_truncate(struct inode *inode)
731 unsigned long blkoff;
732 unsigned int blocksize;
733 struct nilfs_transaction_info ti;
734 struct super_block *sb = inode->i_sb;
735 struct nilfs_inode_info *ii = NILFS_I(inode);
737 if (!test_bit(NILFS_I_BMAP, &ii->i_state))
739 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
742 blocksize = sb->s_blocksize;
743 blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
744 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
746 block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
748 nilfs_truncate_bmap(ii, blkoff);
750 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
752 nilfs_set_transaction_flag(NILFS_TI_SYNC);
754 nilfs_mark_inode_dirty(inode);
755 nilfs_set_file_dirty(inode, 0);
756 nilfs_transaction_commit(sb);
757 /* May construct a logical segment and may fail in sync mode.
758 But truncate has no return value. */
761 static void nilfs_clear_inode(struct inode *inode)
763 struct nilfs_inode_info *ii = NILFS_I(inode);
764 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
767 * Free resources allocated in nilfs_read_inode(), here.
769 BUG_ON(!list_empty(&ii->i_dirty));
773 if (mdi && mdi->mi_palloc_cache)
774 nilfs_palloc_destroy_cache(inode);
776 if (test_bit(NILFS_I_BMAP, &ii->i_state))
777 nilfs_bmap_clear(ii->i_bmap);
779 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
781 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
782 nilfs_put_root(ii->i_root);
785 void nilfs_evict_inode(struct inode *inode)
787 struct nilfs_transaction_info ti;
788 struct super_block *sb = inode->i_sb;
789 struct nilfs_inode_info *ii = NILFS_I(inode);
792 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
793 truncate_inode_pages_final(&inode->i_data);
795 nilfs_clear_inode(inode);
798 nilfs_transaction_begin(sb, &ti, 0); /* never fails */
800 truncate_inode_pages_final(&inode->i_data);
802 /* TODO: some of the following operations may fail. */
803 nilfs_truncate_bmap(ii, 0);
804 nilfs_mark_inode_dirty(inode);
807 ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
809 atomic64_dec(&ii->i_root->inodes_count);
811 nilfs_clear_inode(inode);
814 nilfs_set_transaction_flag(NILFS_TI_SYNC);
815 nilfs_transaction_commit(sb);
816 /* May construct a logical segment and may fail in sync mode.
817 But delete_inode has no return value. */
820 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
822 struct nilfs_transaction_info ti;
823 struct inode *inode = dentry->d_inode;
824 struct super_block *sb = inode->i_sb;
827 err = inode_change_ok(inode, iattr);
831 err = nilfs_transaction_begin(sb, &ti, 0);
835 if ((iattr->ia_valid & ATTR_SIZE) &&
836 iattr->ia_size != i_size_read(inode)) {
837 inode_dio_wait(inode);
838 truncate_setsize(inode, iattr->ia_size);
839 nilfs_truncate(inode);
842 setattr_copy(inode, iattr);
843 mark_inode_dirty(inode);
845 if (iattr->ia_valid & ATTR_MODE) {
846 err = nilfs_acl_chmod(inode);
851 return nilfs_transaction_commit(sb);
854 nilfs_transaction_abort(sb);
858 int nilfs_permission(struct inode *inode, int mask)
860 struct nilfs_root *root = NILFS_I(inode)->i_root;
861 if ((mask & MAY_WRITE) && root &&
862 root->cno != NILFS_CPTREE_CURRENT_CNO)
863 return -EROFS; /* snapshot is not writable */
865 return generic_permission(inode, mask);
868 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
870 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
871 struct nilfs_inode_info *ii = NILFS_I(inode);
874 spin_lock(&nilfs->ns_inode_lock);
875 if (ii->i_bh == NULL) {
876 spin_unlock(&nilfs->ns_inode_lock);
877 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
881 spin_lock(&nilfs->ns_inode_lock);
882 if (ii->i_bh == NULL)
892 spin_unlock(&nilfs->ns_inode_lock);
896 int nilfs_inode_dirty(struct inode *inode)
898 struct nilfs_inode_info *ii = NILFS_I(inode);
899 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
902 if (!list_empty(&ii->i_dirty)) {
903 spin_lock(&nilfs->ns_inode_lock);
904 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
905 test_bit(NILFS_I_BUSY, &ii->i_state);
906 spin_unlock(&nilfs->ns_inode_lock);
911 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
913 struct nilfs_inode_info *ii = NILFS_I(inode);
914 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
916 atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
918 if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
921 spin_lock(&nilfs->ns_inode_lock);
922 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
923 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
924 /* Because this routine may race with nilfs_dispose_list(),
925 we have to check NILFS_I_QUEUED here, too. */
926 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
927 /* This will happen when somebody is freeing
929 nilfs_warning(inode->i_sb, __func__,
930 "cannot get inode (ino=%lu)\n",
932 spin_unlock(&nilfs->ns_inode_lock);
933 return -EINVAL; /* NILFS_I_DIRTY may remain for
936 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
937 set_bit(NILFS_I_QUEUED, &ii->i_state);
939 spin_unlock(&nilfs->ns_inode_lock);
943 int __nilfs_mark_inode_dirty(struct inode *inode, int flags)
945 struct buffer_head *ibh;
948 err = nilfs_load_inode_block(inode, &ibh);
950 nilfs_warning(inode->i_sb, __func__,
951 "failed to reget inode block.\n");
954 nilfs_update_inode(inode, ibh, flags);
955 mark_buffer_dirty(ibh);
956 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
962 * nilfs_dirty_inode - reflect changes on given inode to an inode block.
963 * @inode: inode of the file to be registered.
965 * nilfs_dirty_inode() loads a inode block containing the specified
966 * @inode and copies data from a nilfs_inode to a corresponding inode
967 * entry in the inode block. This operation is excluded from the segment
968 * construction. This function can be called both as a single operation
969 * and as a part of indivisible file operations.
971 void nilfs_dirty_inode(struct inode *inode, int flags)
973 struct nilfs_transaction_info ti;
974 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
976 if (is_bad_inode(inode)) {
977 nilfs_warning(inode->i_sb, __func__,
978 "tried to mark bad_inode dirty. ignored.\n");
983 nilfs_mdt_mark_dirty(inode);
986 nilfs_transaction_begin(inode->i_sb, &ti, 0);
987 __nilfs_mark_inode_dirty(inode, flags);
988 nilfs_transaction_commit(inode->i_sb); /* never fails */
991 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
992 __u64 start, __u64 len)
994 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
995 __u64 logical = 0, phys = 0, size = 0;
998 sector_t blkoff, end_blkoff;
999 sector_t delalloc_blkoff;
1000 unsigned long delalloc_blklen;
1001 unsigned int blkbits = inode->i_blkbits;
1004 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1008 mutex_lock(&inode->i_mutex);
1010 isize = i_size_read(inode);
1012 blkoff = start >> blkbits;
1013 end_blkoff = (start + len - 1) >> blkbits;
1015 delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
1020 unsigned int maxblocks;
1022 if (delalloc_blklen && blkoff == delalloc_blkoff) {
1024 /* End of the current extent */
1025 ret = fiemap_fill_next_extent(
1026 fieinfo, logical, phys, size, flags);
1030 if (blkoff > end_blkoff)
1033 flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
1034 logical = blkoff << blkbits;
1036 size = delalloc_blklen << blkbits;
1038 blkoff = delalloc_blkoff + delalloc_blklen;
1039 delalloc_blklen = nilfs_find_uncommitted_extent(
1040 inode, blkoff, &delalloc_blkoff);
1045 * Limit the number of blocks that we look up so as
1046 * not to get into the next delayed allocation extent.
1048 maxblocks = INT_MAX;
1049 if (delalloc_blklen)
1050 maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1054 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1055 n = nilfs_bmap_lookup_contig(
1056 NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1057 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1062 if (unlikely(n != -ENOENT))
1067 past_eof = ((blkoff << blkbits) >= isize);
1070 /* End of the current extent */
1073 flags |= FIEMAP_EXTENT_LAST;
1075 ret = fiemap_fill_next_extent(
1076 fieinfo, logical, phys, size, flags);
1081 if (blkoff > end_blkoff || past_eof)
1085 if (phys && blkphy << blkbits == phys + size) {
1086 /* The current extent goes on */
1087 size += n << blkbits;
1089 /* Terminate the current extent */
1090 ret = fiemap_fill_next_extent(
1091 fieinfo, logical, phys, size,
1093 if (ret || blkoff > end_blkoff)
1096 /* Start another extent */
1097 flags = FIEMAP_EXTENT_MERGED;
1098 logical = blkoff << blkbits;
1099 phys = blkphy << blkbits;
1100 size = n << blkbits;
1103 /* Start a new extent */
1104 flags = FIEMAP_EXTENT_MERGED;
1105 logical = blkoff << blkbits;
1106 phys = blkphy << blkbits;
1107 size = n << blkbits;
1114 /* If ret is 1 then we just hit the end of the extent array */
1118 mutex_unlock(&inode->i_mutex);